Tara | Payment Sanction Screening Alert Review

Version 4.1.0

About Release

• Release date: June 10, 2026
• Compatible platform versions: Work.AI 10.3+

The release adds support for alert- and hit-level rules, enhances decision reapplication (CLDS), introduces email notifications for quality check (QC) and CLDS reports, and improves message content parsing, reporting, and billing.

• You can now manually disable billing by creating a dedicated pss_disable_billing secret entry in Secrets Vault. This also prevents billing-related error messages from cluttering the logs.

• A new IND (Individual) and ORG (Organization) ownership rule escalates cases where a sanctioned individual's name is fully contained in a screened organization name, helping identify potential ownership or control relationships between individuals and organizations.

• Time-based data collection in QC and CLDS reports is improved:

  • For the Past N days option, current-day records are excluded.

  • For the Custom date range option, a specific time range is now applied.

  • Time zone support is added. By default, the Control Tower UTC time zone is used. All database queries, QC reports, and CLDS reports use and display timestamps in the configured time zone.

• To avoid redundancy and the risk of inconsistencies, the derived object is removed from the rule model. Its data is redistributed to dedicated match objects (idMatch, addressMatch, countryMatch, and dobMatch), which serve as the authoritative sources during rule evaluation. As a result, Tara v4.1.0 is not backward compatible with previous rule files.

  • DOB matching now calculates and sets the match result based on the configured threshold, rather than relying on rule logic.

  • ID matching correctly sets the source (screened vs NER) and ID subtype directly on idMatch, with consistent use of pipe-separated values.

  • Address matching evaluates all available address sources (screened, NER, and geocoder-enriched data) instead of stopping at the first match and sets an address confirmation flag.

  • Country validation invalidates only specific countries that fail geocoder lookup, rather than clearing all country data when a single value is invalid.

  • Split hit entity resolution in the decision engine now uses entity data from match objects instead of the removed derived objects.

  • Rules referencing $hit.derived.*, for example, $hit.derived.subType or $hit.derived.countryConfirmed, are no longer supported and must be updated to use the appropriate match object.

• QC and CLDS report generation capabilities are extended to send report links to specific email addresses. Email notification settings are available on the Output step of the configuration form.

• NER document creation and result processing for multi-tag content are enhanced to prevent false escalations caused by incorrectly merged entities:

  • Tag content is deduplicated during document creation.

  • Entities of the same type are no longer merged across multipart tags during NER result processing.

• The rules framework offers clearer structure, flexibility, and greater control over decision outcomes:

  • Hit-level and alert-level rules are now separated into the hit_rules.drl and alert_rules.drl files, each with its own agenda group ("hit" and "alert"), ensuring better organization, predictable execution, and easier troubleshooting.

  • Alert-level decision comments are defined in alert_rules.drl, with the getStatisticalAlertLevelComment() method used to generate them. Alert rules populate the MessageDecision comment using the DecisionStatistic object.

  • In addition to standard decision statuses like RESOLVE and NO_DECISION, custom statuses, for example, PASS, HALF_PASS, L1, or L2, are supported for both HitDecision and MessageDecision.

  • You can now define custom hit and alert rule files in the Model step of the configuration form. When provided, these custom files replace the default rules, so you have full control over rule logic and decision outcomes.

• For easier understanding and maintenance, the Enrichment object is split into LocationEnrichment (external address and geolocation data) and EntityEnrichment (security and entity-related data). Backward compatibility is preserved as existing data is automatically converted with no impact on results.

• To reduce the risk of unintentionally releasing relevant hits, the decision reapplication provides enhanced blacklisted country detection. When enabled, hits containing a blacklisted country are no longer automatically released, and newly added sanctioned countries are consistently enforced across both new and previously processed data.

• CLDS now adapts based on recent analyst decisions, so outcomes are more accurate and consistent. When Update CLDS status based on most recent analyst decisions is enabled, Tara evaluates recent analyst actions to dynamically adjust how CLDS handles hits. Repeated Release and Resolve decisions can allow CLDS to release a hit even if earlier decisions prevent it, while repeated Escalate decisions can block automatic release. The Last N analyst Release/Resolve and Last N analyst Escalate configuration fields control this behavior.

• The Swift ADK parser is updated to handle message content where the same tag appears multiple times, which results in more accurate data extraction and fewer gaps in downstream processing.

• The FUF message content parser now supports partial tag matching. It identifies and extracts tag values even when tag names differ only by a numeric suffix, for example, ABC001 and ABC002. If no exact match is found, the parser attempts a partial match and, for multiple matches, combines values into a single result. The enhanced logic ensures more reliable tag extraction for both single and multiple tag occurrences and supports downstream processing such as NER.

• The message content parser is enhanced to identify tag content based on hit text when tag names are missing or do not match, ensuring more reliable data retrieval. This fallback mechanism applies when a single matching tag is found, enabling reliable parsing even with incomplete or inconsistent message formats.

  The parser also improves detection of sanctioned countries and cities in name fields by normalizing and cleaning name values and inferring missing location data where possible.

• The message type in the input contract is now optional and uses a String type instead of an Enum, reducing dependency on predefined values and simplifying integration with client data. Supported values include SWF and ISO_20022; others are deprecated. For backward compatibility, a default value of PAYMENT_TRANSFER can be used where needed.

• Billing Service integration ensures more consistent and accurate data handling. The service call uses the Java Native Worker BillingInfo object instead of BillingServiceInfo, aligning with the standardized data model. The Business Process variation title is now retrieved via controlTowerBpService.

• The AI Agent configuration is upgraded with the Decision model v6.1.0 and Name Matcher v2.0.1 and includes the following improvements:

  • Refined matching scores. 1.0 is returned only for exact matches after preprocessing, for example, "John Patrick Smith" vs "John Patrick Smith".

  • Normalized "Last, First [Middle]" name formats. Names containing a comma are converted into the standard order before comparison, for example, "Jackson, Corey A." → "Corey A. Jackson".

  • Add non-equivalent name pairs. Pairs such as "amin" and "amir" are included to prevent false positives.

  • Extend the organization stop-word list. Terms such as "groups" are added to improve organization name matching.

  • Expanded candidates for the MD prefix. Additional variants are generated by replacing "MD" with "Mohammed" and both options are evaluated.

  • Applied token coverage penalties. Scores are reduced when a watchlist name is embedded within a much longer screened name.

  • Adjusted scoring for name-in-company patterns. A mild penalty is applied when a watchlist name matches only part of a longer organization name.

  • Improved stop-word handling in preprocessing. Word boundaries are enforced to avoid matching stop words inside larger terms.

  • Restricted country code removal. Removal is applied only when the code appears at the end of the name.

  • Refined number-word removal. Removal is skipped when it would leave only a very short token (≤ 2 characters).

  • Corrected handling of legal-form abbreviations. Trailing name parts are preserved when company identifiers like "Corp" or "Ltd" are present.

  • Updated Python package import paths by moving modules from wf_name_matcher.matching.* to wf_name_matcher.domain.matching.*.

  • Added suffix matching. Matches are recognized where one token is a suffix of another, for example, "Gazprom" vs "Belgazprom".

  • Reduced acronym-based false positives. Acronym letters must match the order of the corresponding words.

  • Fixed scoring for near-identical organization tokens. Similarity rules are now used instead of treating such tokens as abbreviations.

  • Corrected single-word matching behavior. Over-penalizing matches are avoided against longer names.

  • Improved stop-word-only matching logic by reducing false positives when only common terms match with high confidence.

  • Resolved edge cases where no high-confidence token pairs are found during stop-word-only matching.

• The NER model is upgraded to v3.2.6 with the following changes:

  • Improved person name recognition. Titles and honorifics, for example, Dr, Prof, or MD, are now included as part of the name. For example, "Dr. Jane Smith" is extracted as a single entity.

  • Ensured accurate confidence scores for organizations. Organization entities now reflect the model's actual confidence, providing consistent scoring across entity types.

  • Updated regex code for ID_LEI to ensure both alphabetic and numeric characters are present.

  • Fixed the preprocessing regex issue that caused incorrect splits inside UUIDs.

  • Introduced Java 8 compatibility. The model is now available as a dedicated Java 8 bundle alongside Java 21.

  • Improved entity recognition in specific scenarios:

    • German e-commerce payment references to avoid duplicate organization detection

    • SWIFT indicator tag content to ensure classification as a named entity

    • Mixed entity blocks in SWIFT fields to separate person, organization, and address correctly

  • Corrected confidence scores for indicator-tagged entities. Actual model probabilities are now used instead of inflated values.

  • Improved vessel name detection. Structural prefixes, for example, MT or IMO, are no longer included as part of the vessel name.

  • Improved recognition of financial organizations in securities payment messages. Entities like BLACKROCK and GOLDMAN SACHS are now correctly extracted as ORGANIZATION when surrounded by financial identifiers.

  • Improved handling of "trading as" patterns. For example, "DANIEL SMITH TRADING AS CONSULTING MARKETING" correctly returns both PERSON and ORGANIZATION.

  • Improved recognition of romanized Chinese personal names and addresses.

  • Fixed the entity offset issue by correcting begin positions for salutation-prefixed names, for example, "MRS JANE SMITH".

  • Removed invalid entities. Punctuation-only outputs, such as a lone comma, are suppressed.

Version 4.0.3

About Release

• Release date: April 3, 2026
• Compatible platform versions: Work.AI 10.3+

The Tara 4.0.3 release focuses on improving rule inputs, matching quality, data processing resilience, and model behavior. It also introduces updates to billing events, historical data upload, and adjudication logic.

• The rule object now includes all processed sanctioned entity names, including the watchlist entity (WLE) name, primary name, and alias, along with their corresponding name matcher scores, enabling more precise custom rule authoring to increase the automation rate.

• To prevent individual rule failures from interrupting Business Process execution, runtime exceptions during rule processing are now caught at the hit level, returning a NO_DECISION outcome with a failure narrative.

• A new Rare and Screened token overlap mode gives you stricter control over name matching and helps reduce false positives in high-risk workflows. The new Rare and Screened option is available in the Token overlap configuration field in the configuration form and is propagated through the Business Process to the Decision Model.

• Billing events now contain additional AI Agent metadata, including the AI Agent variation, name, and skill, which improves billing accuracy and helps ensure compliance with billing requirements.

• To make historical data ingestion more flexible and reduce failures caused by missing fields, continuous learning (CLDS) historical data upload provides improved support for Firco v6 and similar screening systems:

  • Added optional hit-level analyst decision fields (T_IS_RELEASED and T_DECISION_TYPE), so the upload process can preserve more granular review outcomes.

  • Added optional hit-level tag content (T_TAG_CONTENT), while still allowing Tara to derive tag content from the message when available.

  • Made the T_MESSAGE, T_NATURE, and T_CONTENT_FORMAT columns optional, so missing or blank values in these fields no longer cause a Business Process to fail.

• Payment information, such as payment amount, currency, and transfer direction, is now extracted from FUF and SWIFT content, enriching alert context and enabling more accurate rule-based adjudication.

• Data cleanup removes commonly used placeholder values from alert data during both CLDS historical data upload and real-time processing to prevent placeholder values from interfering with CLDS compatibility and match accuracy. By default, the cleanup logic deletes values such as Unknown, N/A, and *, replacing fully matched placeholder content with empty values before further processing.

• When the tag content is a substring of the screened address, Named Entity Recognition (NER) processing uses the full address as tag content to provide richer context and improve location entity prediction accuracy. The enhancement is designed for cases where the original tag content is embedded in a longer address and would otherwise not provide enough context for reliable location detection.

• To reduce incorrect location validation caused by partial or misleading address matches, geocoder lookup now contains two configurable location validation modes. The Strict option marks country and address as valid only when the full address is confirmed, whereas Approximate allows country validation for partial area-level matches.

• The rule object provides more complete and accurate names and name matcher scores, resulting in more reliable inputs for custom rule design and adjudication:

  • Added all available name matcher similarity scores for screened, lookup, and NER data compared against sanctioned entity names, primary names, and aliases.

  • Restored original token spacing when processed names are concatenated (for example, "lesterkub" → "Lester Kub"), which makes matched names easier to interpret in rules.

  • Changed score and name selection to follow input source preference (screened data first, then lookup data, then NER data) rather than the highest score, ensuring more consistent rule inputs.

• Synthetic data generation has been updated to v34, incorporating refactored generation methods and updated screening scenarios.

• Adjudication rules cover more client-specific scenarios to improve adjudication accuracy and increase the automation rate:

  • Used the NER source type when replacing the NAME type.

  • Added new textContainsAnyWleNames(), isNameInTheMiddle(), getMessageContentTagValue(), and isNamesEqual() rule functions to support more flexible rule logic.

  • Refined source handling and matched WLE logic to treat names as primary names when primary names are absent.

  • Added logic to update the sanctioned code map when a sanctioned name resembles an ID, improving downstream rule evaluation in edge cases.

  • Enhanced rule-processing stability by preventing null pointer exceptions on empty decision reasons when building a decision object.

• To prevent unbounded table growth in production environments, the pss_reapplication_hashcodes_v4 Data Store includes the decision_date column and supports configurable data purge to automatically remove records older than the defined retention period.

• Hit entity prediction delivers more accurate and explainable results, reducing incorrect hit text selection and unnecessary ambiguity in alert analysis:

  • Replaced the fuzzy matching score of hit text with a character matching ratio to improve the precision of direct text comparisons.

  • Introduced a preprocessing step that extends partially matched hit text to the nearest token in tag content to better identify direct matches.

  • Reduced unnecessary text splits by targeting specific hit text boundary handling and filtering ambiguous words.

  • Reduced the number of unknown hit entities, so this outcome occurs only when the compared names have no meaningful overlap.

• The Decision Model is upgraded to v6.0.9 with Name Matcher v1.2.3 and includes the following improvements:

  • Fixed an unhandled exception for the AN/ANT (Netherlands Antilles) and CS/SCG (Serbia and Montenegro) input country codes. Legacy country codes that were later split are now converted to the corresponding new country codes (for example, SCG → SRB and MNE) and matched against any matching parts in the results.

  • Fixed the issue where aggregated_features names were blank when screenedData.name was not provided. Processed WLE names can now be taken from nerData or securityData.

  • Added original and processed names from screenedData, securityData, and nerData to the output for use in rules.

  • Updated libpostal output to use the ISO country code and city extracted from the identified SWIFT address.

  • Enhanced organization matching for names with legal forms in different positions. For example, SUEKINO AG now matches INTERNATIONAL LIMITED LIABILITY COMPANY SUEKINO LTD.

  • Improved legal form handling by removing consecutive legal form abbreviations (for example, Company LLC Inc).

  • Improved handling of duplicate legal forms. Names such as Company LLC Holdings LLC trigger truncation at the first occurrence for cleaner matching.

  • Preserved legal form abbreviations followed by ambiguous company terms, such as AB Group or KS Holdings, to avoid overstripping meaningful name parts.

  • Improved handling of legal forms at the beginning of names to prevent incorrect text removal.

  • Enhanced organization matching for quoted company names within legal form wrappers. For example, International Joint Stock Company 'NAME Holdings Limited' and NAME HOLDINGS LIMITED are now treated as the same entity.

  • Improved handling of names with apostrophes and similar characters. Names such as O'Brien correctly match O Brien by normalizing quote-like characters to spaces.

  • Added support for alias indicators in names. Names containing phrases like t/as, trading as, aka, also known as, dba, doing business as, formerly, or f/k/a are automatically split, and both parts are matched separately. For example, ABC Corp t/as XYZ Services matches against both ABC Corp and XYZ Services.

  • Enhanced legal form comparison to properly handle organization names with multiple legal forms. Previously, only single legal forms were compared. All combinations are now checked for equivalence.

• The NER model is upgraded to v3.2.2 with the following improvements:

  • Enhanced securities, fund, and stock name recognition through targeted model fine-tuning to reduce false positives and incorrect NER predictions.

  • Improved organization detection by preventing ISO country codes, such as AE and AG, from being treated as business entity legal forms, reducing false ORGANIZATION classifications for LOCATION and PERSON entities.

  • Refined organization merger logic to skip merging adjacent entities when the leading legal form matches an ISO code, preserving correct entity types in mixed address and company text.

  • Improved entity classification accuracy for entities containing location markers (for example, street types, address abbreviations, known city, or region names) even when they are initially misclassified. This reduces false positives in text with many addresses.

  • Expanded detection of location-like patterns across both ORGANIZATION and PERSON entities for more reliable disambiguation between people, companies, and addresses.

  • Updated reference data to improve recognition accuracy.

Bug fixes

• Resolved the issue where the reason narrative for a single merged hit incorrectly included the SPLIT marker.

• Fixed the null pointer exception when required data was missing during time difference calculation in the PSS QC Report Business Process, which prevented the quality control report from being generated.

• Resolved the issue when a Business Process failure was routed to the Apply rules step while the rule object was missing, causing processing to get stuck.

• Fixed the null pointer exception in DroolsService that occurred when creating a container with null rules content.

Version 4.0.2

About Release

• Release date: January 20, 2026
• Compatible platform versions: Work.AI 10.3+

The Tara 4.0.2 release addresses licensing compliance by replacing a GPL-licensed component and delivers improvements to name matching, hit detection logic, and rule processing.

• The previous GPL-licensed fuzzy matching library has been replaced with a license-compliant fuzzy scoring algorithm, eliminating legal and distribution risks while preserving existing matching behavior and result quality.

• To improve geographic matching accuracy and strengthen risk detection, new getCountriesFromText() and isTextCountryMatchesWleCountry() rule functions enable the extraction of countries from free text and comparison against sanctioned entity countries.

• Tara now ignores salutations (for example, Mr., Dr.) or correctly merges them with personal names, preventing false hit splits and enhancing match accuracy.

• Overlap results that are only one character, digits only, or trivial two-character merges are now excluded. The reworked hit overlap filtering reduces noise in matching results and improves overall screening precision.

• Only unique sanctioned names are now sent to the name-matching engine, preventing aliases from being incorrectly selected over primary names. The enhancement ensures consistent, compliant narratives and reduces the risk of misleading match explanations.

• Null values in string types are now normalized to empty strings before rule execution, preventing runtime exceptions and strengthening processing stability.

• The NER model is upgraded to v3.0.7 with the following improvements:

  • Fixed the text preprocessing decoding issue where unexpected punctuation could cause an unhandled UnicodeDecodeError: 'unicodeescape' codec can't decode exception, resulting in empty entities_found output.

  • Improved organization entity recognition by updating the list of known company legal suffixes.

• The configuration is upgraded with the Decision model v6.0.2, resolving the issue where the Name Matcher could select an alias with the same tokens as sanctionData.name or sanctionData.primaryName.

Version 4.0.1

About Release

• Release date: December 22, 2025
• Compatible platform versions: Work.AI 10.3+

The Tara 4.0.1 release focuses on strengthening security, improving screening accuracy, and refining AI Agent's configuration and enrichment performance.

• Java vulnerabilities identified by Snyk and JFrog scans are addressed. The updates reduce security risk and improve deployment reliability.

• The Tara UI configuration screens now match other AI Agent configuration screens, with updated presentation and wording to deliver a more consistent, clear, and user-friendly configuration experience.

• The SWIFT parser now supports multiple identifiers in the 35B field, extracting all available IDs from the tag content to improve data completeness and reduce manual intervention during message processing.

• The name-based search accuracy within the GLEIF API is improved by accounting for common spelling variations such as extra or missing spaces, omitted stop words, and incorrect punctuation, which results in more reliable matches and fewer missed enrichments.

• Tara now supports updated MLSDK model output formats, ensuring compatibility with newer MLSDK versions by handling both object-based and string-based results for the NER and Decision Models and preventing processing issues after model upgrades.

• The JNW Generic Connector is upgraded to v3.6.0 for the Orbis connector, fixing vulnerabilities.

• Secrets Vault entries can now be created directly from the configuration form, enabling users to add and automatically apply new API credentials without leaving the setup flow, streamlining configuration and reducing setup time.

• Enrichment connectors are enabled to cache negative lookup results, storing non-conclusive responses from Orbis and GLEIF with an expiration period to avoid repeated unnecessary lookups and optimize data retrieval processes.

• The GLEIF connector now refreshes cached lookup data automatically, using a configurable cache retention period to keep data up to date while reducing unnecessary repeated lookups, similar to the existing logic in the Orbis connector.

• Tara now uses the sanctioned name and type as the screened name and type when tag content fully matches a sanctioned entity, improving screening accuracy and consistency while preserving existing screened data.

• Hidden Data Store fields are removed from the AI Agent JSON configuration. Hard-coded, version-bound Data Store values are used instead to reduce runtime lookups, prevent configuration-related failures, and improve overall processing stability.

• The AI Agent configuration is upgraded with the Decision model v6.0.1, introducing the following enhancements:

  • The Java hypermodel worker now requires Java 21, ensuring compatibility with updated runtime dependencies.

  • The model output contract is updated, with attrs.model-results provided as a JSON string instead of an object.

  • Support for caching negative enrichment outcomes is added, reducing repeated lookups and improving overall processing efficiency.

  • Dependencies and component versions are updated to enhance stability and maintainability.

• The NER model is upgraded to v3.0.5 with the following improvements:

  • The PersonValidation post-processor reduces confidence for ambiguous PERSON entities, such as multiple names joined by connectors, for example, and/or or und/oder.

  • Person name lookup support is added to the configuration, improving the accuracy of person recognition rules.

  • Overconfident ORGANIZATION entities are further reduced by identifying and handling location-only markers using standard abbreviations and known location names.

  • The asterisk character (*) is added to the preprocessing replacement list to ensure consistent text normalization.

  • Python dependencies are updated and cleaned up, resolving known vulnerabilities.

  • The Python environment layout is updated. The initial deployment requires the bundled environment artifact (bundle-with-env), while subsequent deployments can reuse the existing environment.

  • The OrganizationPrefixSplitter post-processor is introduced to split leading reference prefixes (for example, REF and REF TL) from ORGANIZATION entities, producing cleaner company names.

  • Additional financial instrument ID types are supported, including SEDOL and FIGI, with improved prediction and validation for CUSIP, FIGI, ISIN, and SEDOL.

  • The new entity type ID_IMAD is introduced, supporting IMAD and OMAD identifiers used in FedWire payments for tracking and reconciliation.

  • BIC detection logic is revised to reduce false positives where personal names were previously misclassified.

  • The list of known companies is updated, improving organization recognition accuracy.

  • NER processing performance is optimized for duplicate tag content. Identical text is now processed only once and reused across hits to reduce processing time.

Bug fixes

• Resolved the issue with a quality check (QC) report not reflecting security data usage in the screened name. When a security name is used for name matching, the QC report now shows the screened security name. In addition, the data source column now contains the LOOKUP type, making it easier for users to understand how the data was sourced.

• Resolved the data purge issue where archival data purge scripts caused bundle import failures due to a Liquibase migration error. The fix results in smoother upgrades and more reliable data management.

• Fixed the typeMatch logic in the rules JSON object. When the type is set to location, address and country values are displayed instead of the name. If an address is not available, other relevant location details are shown, ensuring more meaningful and reliable matching results.

• Resolved the issue where GLEIF was not selected by default on the Data Enrichment configuration form when enabling the use of connectors' data. The fix ensures a more intuitive setup and reduces the risk of incorrect configuration.

• Fixed the issue with SWIFT tag content line splitting where lines starting with SWIFT markers (for example, 1/, 2/) were not correctly rejoined. The fix improves NER model output by accurately reconstructing text and ensuring correct identification and extraction of IDs, organization names, and locations.

• Fixed the issue in hit text calculation where hits on identifiers with a one-character prefix (for example, I/HU and O/US) were incorrectly attributed to a location value. The logic now correctly identifies and presents the intended ID as the hit for adjudication, aligning results with user expectations.

• Resolved the issue where small IDs starting with / (for example, /0B7Z in SWIFT tag 59F) were not recognized as identifiers. The fix enhances ID detection and matching accuracy while avoiding false positives in cases like /Name1/ABC, where Name1 is not an ID.

• Fixed the issue where ISINs were incorrectly split during NER preprocessing (for example, CNE100004SP7 becoming CNE and 100004SP7), preserving valid ISIN tokens end-to-end and preventing entity misclassification.

Version 4.0.0

About Release

• Release date: October 17, 2025
• Compatible platform versions: Work.AI 10.3+

The release modernizes Tara to be compatible with Work.AI v10.3, enhancing performance, maintainability, and scalability. It also introduces real-time analytics and strengthens data management and security.

• To align with Work.AI 10.3 dependencies, the Tara codebase is migrated to Java 21, with all third-party libraries upgraded to the latest compatible versions for enhanced stability and runtime efficiency.

• All Data Stores are updated with the _v4 suffix to facilitate more efficient data maintenance. Unused Data Stores are deleted to streamline platform resources and reduce maintenance overhead.

• To improve database version control, auditability, and deployment consistency across environments, a new set of Liquibase scripts is created for each Data Store.

• Rather than relying on batch processing, Tara now streams events in real time through RabbitMQ. The Payment sanction screening overview dashboard is updated to consume these live events directly from the analytics service instead of Data Stores, providing faster operational insights, greater data accuracy, and more responsive decision-making.

• Data purge and archiving functionality is now included in the bundle, with purges disabled by default and separated between analytics and request processing. Purge frequency and data retention periods are adjustable, with the default retention period set to three months. Now, each Data Store can follow its own retention policy, which helps to maintain optimal storage performance and compliance with data governance standards.

• The JNW Generic Connector is upgraded to v3.5.0 for the Orbis Enrichment Connector, ensuring improved data quality during enrichment processes.

Bug fixes

• Resolved Snyk security vulnerabilities by updating multiple third-party dependencies and upgrading the platform components to the latest version.

Version 3.4.3

About Release

• Release date: February 17, 2026
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.4.3 release improves matching and entity recognition, updates synthetic data generation, and provides upgraded models and stability fixes.

• The previous GPL-licensed fuzzy matching library has been replaced with a license-compliant fuzzy scoring algorithm to eliminate legal and distribution risks while preserving existing matching behavior and result quality.

• Tara now ignores salutations (for example, Mr., Dr.) or correctly merges them with personal names, preventing false hit splits and enhancing match accuracy.

• Synthetic data generation is updated to v32 by removing unstable scenarios and ensuring hit text partially matches tag content to prevent random hit text selection.

• The NER model is upgraded to v2.9.0 with the following improvements:

  • Fixed the text preprocessing decoding issue where unexpected punctuation could cause an unhandled UnicodeDecodeError: 'unicodeescape' codec can't decode exception, resulting in empty entities_found output.

  • Improved organization entity recognition by updating the list of known company legal suffixes.

  • Enhanced security, fund, and stock name recognition through targeted model fine-tuning.

  • Refined named-entity recognition accuracy by reducing incorrect organization labels on personal names. Names are now classified as people where expected.

• The AI Agent configuration is upgraded with the Decision model v5.2.7 and Name Matcher v1.2.2 that include the following changes:

  • Fixed enrichment connector lookups by saving a negative (non-conclusive) security record with an expiration to reduce repeated calls for missing or undetermined results.

  • Fixed an unhandled exception for the AN/ANT (Netherlands Antilles) and CS/SCG (Serbia and Montenegro) input country codes. Legacy country codes that were later split are now converted to the corresponding new codes (for example, SCG → SRB and MNE) and matched against the countries in the results.

  • Removed the patch that suppressed the matchIfRareTokenOverlap setting. Name Matcher now adheres to the AI Agent configuration. The new Rare and Screened token overlap setting combines rare-token and screened-token overlap checks for stricter matching and fewer false positives in high-risk workflows.

Bug fixes

• Fixed a null pointer exception when required data was missing during time difference calculation in the PSS QC Report Business Process, which prevented the quality control report from being generated.

• Resolved the issue when a Business Process failure was routed to the Apply rules step, but the rule object was missing, causing processing to get stuck.

• Fixed the null pointer exception in DroolsService that occurred when creating a container with null rules content.

Version 3.4.2

About Release

• Release date: December 24, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.4.2 release delivers improvements to enrichment accuracy, screening reliability, and connector stability.

• The SWIFT parser now supports multiple identifiers in the 35B field, extracting all available IDs from the tag content to improve data completeness and reduce manual intervention during message processing.

• The JNW Generic Connector is upgraded to v2.9.2 for the Orbis connector, fixing vulnerabilities.

• The name-based search accuracy within the GLEIF API is improved by accounting for common spelling variations such as extra or missing spaces, omitted stop words, and incorrect punctuation, which results in more reliable matches and fewer missed enrichments.

• Enrichment connectors now cache negative lookup results, storing non-conclusive responses from Orbis and GLEIF with an expiration period to avoid repeated unnecessary lookups and optimize data retrieval processes.

• The GLEIF connector now refreshes cached lookup data automatically, using a configurable cache retention period to keep data up to date while reducing unnecessary repeated lookups, similar to the existing logic in the Orbis connector.

• The AI Agent configuration is upgraded with the Decision model v5.2.1, introducing support for caching negative enrichment outcomes to reduce repeated lookups and improve overall processing efficiency.

• The NER model is upgraded to v2.7.5 with the following improvements:

  • The PersonValidation post-processor reduces confidence for ambiguous PERSON entities, such as multiple names joined by connectors, for example, and/or or und/oder.

  • Person name lookup support is added to the configuration, improving the accuracy of person recognition rules.

  • Overconfident ORGANIZATION entities are further reduced by identifying and handling location-only markers using standard abbreviations and known location names.

  • The asterisk character (*) is added to the preprocessing replacement list to ensure consistent text normalization.

  • Python dependencies are updated and cleaned up, resolving known vulnerabilities.

  • The Python environment layout is updated. The initial deployment requires the bundled environment artifact (bundle-with-env), while subsequent deployments can reuse the existing environment.

  • The OrganizationPrefixSplitter post-processor is introduced to split leading reference prefixes (for example, REF and REF TL) from ORGANIZATION entities, producing cleaner company names.

  • BIC detection logic is revised to reduce false positives where personal names were previously misclassified.

  • The list of known companies is updated, improving organization recognition accuracy.

Bug fixes

• Resolved the issue with a quality check (QC) report not reflecting security data usage in the screened name. When a security name is used for name matching, the QC report now shows the screened security name. In addition, the data source column now contains the LOOKUP type, making it easier for users to understand how the data was sourced.

• Fixed the typeMatch logic in the rules JSON object. When the type is set to location, address and country values are displayed instead of the name. If an address is not available, other relevant location details are shown, ensuring more meaningful and reliable matching results.

• Resolved the issue where GLEIF was not selected by default on the Data Enrichment configuration form when enabling the use of connectors' data. The fix ensures a more intuitive setup and reduces the risk of incorrect configuration.

• Fixed the issue with SWIFT tag content line splitting where lines starting with SWIFT markers (for example, 1/, 2/) were not correctly rejoined. The fix improves NER model output by accurately reconstructing text and ensuring correct identification and extraction of IDs, organization names, and locations.

• Resolved the issue where small IDs starting with / (for example, /0B7Z in SWIFT tag 59F) were not recognized as identifiers. The fix enhances ID detection and matching accuracy while avoiding false positives in cases like /Name1/ABC, where Name1 is not an ID.

• Fixed the issue where ISINs were incorrectly split during NER preprocessing (for example, CNE100004SP7 becoming CNE and 100004SP7), preserving valid ISIN tokens end-to-end and preventing entity misclassification.

Version 3.4.1

About Release

• Release date: October 17, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.4.1 release enhances the data enrichment, entity recognition, and decision intelligence capabilities to strengthen data validation, reduce processing errors, and deliver more consistent decision outcomes.

• The GLEIF connector is added as a separate step to the PSS Enrichment Connectors Business Process for entity information lookup, supporting entity name validation and address enrichment. The new service allows a configurable search for company information using security identifiers such as LEI, ISIN, or BIC, with multi-ID batch calls and caching by source.

• The sanctions-api-client is upgraded to v1.2.10. The SecurityInfo object now includes additional fields from GLEIF enrichment, which enables richer entity profiles and improved compliance checks.

• The AI Agent configuration is upgraded with the Decision model v5.2.0, introducing the following improvements:

  • Added support for the SecurityData object from enrichment data to extract names for enhanced name-matching accuracy.

  • Improved the decision-making process, resolving the issue where the model returned inconsistent best-matched aliases for the same input. When multiple names on one side produce the same matching score, Tara now sorts names by length to ensure deterministic and reliable outputs.

• The NER model is upgraded to v2.6.6, delivering higher precision in entity extraction and improved model robustness through the following updates:

  • Enhanced recognition of ISO codes as LOCATION entities.

  • Improved ID handling by excluding specific false-positive patterns.

  • Refined quote and punctuation preprocessing for cleaner tokenization.

  • Enhanced preprocessing to split alpha prefixes from numeric suffixes when the suffix contains [./-].

  • Improved validation when, on invalid input, the model returns a single OUTSIDE entity (confidence 0.1) and logs a warning to ensure traceability.

  • Improved location detection by reducing confidence when no digits are present and every word has a long consonant sequence (more than five consonants).

  • Refined organization detection:

    • Organization name merges are only allowed when the second token is a pure legal or business indicator.

    • The confidence for organization-like names that resemble locations (for example, those containing strong street, avenue, or parkway markers) is reduced to minimize false positives in mixed-address text.

  • Cleaner tokenization around the word BANK, ensuring that real words like banks, banking, banker(s), and bankruptcy are no longer split, while concatenated names (for example, BANKOFAMERICA → BANK OF AMERICA) are still separated for better recognition and consistency.

  • Minor stability and configuration updates.

Bug fixes

• Fixed null pointer exceptions during date conversion in message parsing, which previously caused incomplete records and quality control report failures when some data fetched from the pss_processing Data Store lacked decisions.

• Resolved the issue where Orbis query results occasionally returned null or missing values.

• Resolved the issue with NER input truncation for multi-tag content beyond the 35th character. Tags with the same name (without numerical extension, such as "[ABC]") and values shorter than 36 characters are now joined into full text, improving data integrity and recognition accuracy.

Version 3.4.0

About Release

• Release date: August 29, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.4.0 release focuses on extended Named Entity Recognition (NER) and Decision model capabilities, improved process integration, enhanced data enrichment for compliance workflows, and stronger performance and stability.

• Tag list dictionaries for FUF, ISO, and SWIFT are moved from Data Stores to CSV files in S3 (payment_sanctions_screening/{version}/tag-dictionary), eliminating dependencies on mutable Data Stores.

• To improve handling of special characters in text preprocessing, single quotes are replaced with safe delimiters. The approach resolves misclassification issues, helps avoid tokenization issues, and enhances recognition of Arabic and transliterated names.

• Support for the NER additional output possible_types field is added to cover VESSEL recognition from preset known VESSEL names. The change updates the NEREntity object in schema.yaml, rules to account for additional NER possible_types, and rules for the input NAME type.

• Support for subprocess execution of Tara Business Process is added, with clearly defined input and output contracts to simplify integration. The AI Agent now returns rest_response and pss_response to the caller. This is made possible by:

  • Adding the Message object pss_request to the input contract and the MessageProcessingResponse object pss_response to the output contract

  • Adding html_report_link to the MessageProcessingResponse object

  • Deleting unused connector_request, connector_response, and send_request options

  • Moving all JSON objects between steps as objects, not strings

• The Address Verification service is updated to support a new output contract that can be used by all AI Agents, enabling broader reusability across workflows.

• The logic of removing the 36th new line character is updated, including the following changes:

  • Fixed issue with the Windows style \r\n line breakes

  • Fixed issue with trailing spaces before the 36th character

  • Additional English dictionary to check words before deleting the 36th new line character

• The ability to compare against CSV data files in the Drools rule processing engine is added. CSV data is now downloaded from the S3 doc-upload bucket. You can compare alert and hit fields against CSV rows using CSV cache helper functions. Relevant sample rules are added to the custom_rules_sample.drl file to enable more flexible and data-driven rule management.

• The SWIFT parser is updated for tag 35B and SWF_4_35B to extract ISIN, SEDOL, CUSIP, or FIGI security IDs and country code, improving financial instrument recognition.

• The custom Orbis Connector is added for security lookup:

  • The Data Enrichment configuration form is updated to support Orbis Connector.

  • The Business Process flow includes new enrichment connectors as the PSS Enrichment Connectors subprocess.

  • Sample rules are added to the custom_rules_sample.drl file.

• The sanctions-api-client is upgraded to v1.2.9, with the following objects added:

  • SecurityInfo object in EnrichmentInfo to hold security data lookups

  • More fields in the PaymentOrigin object

  • alternateNERTypes list in NEREntityInfo to hold additional possible types

• The NER model is upgraded to v2.6.3 with the following improvements:

  • Vessel database validation and enhanced vessel entity logic

  • Better model predictions for structured text with empty name tags

  • Resolved issue with removing hyphens in names during preprocessing

  • Improved recognition of transliterated Chinese addresses, individuals, and companies

  • Improved handling of single quotes

  • Validation for excessive commas relative to token count

  • Ehanced single quote handling to remove quotes instead of replacing with spaces

  • Extended output contracts with the possible_types field

  • Improved vessel entity processing with the possible_types field

  • Prevention of false positives when input is just a SWIFT Service Identifier (SSI) without context

  • Better separation of locations from company or financial terms

  • Improved recognition of full company legal forms and variants

  • More reliable merging of organization names split by numbers or punctuation

  • Expanded reference data for business entities and locations

  • Updated lookups

  • Improved handling of SSI-like strings, slashes, hyphens, and newlines

  • Filtering of financial terms or company names mislabeled as locations

  • Performance and stability improvements from shared resource loading and caching

  • Reduced false positives for non-entities, such as payment purposes, greetings, or generic requests

  • Improved organization parsing, for example, merging legal forms appearing first

  • More accurate results in mixed financial and address text

• The AI Agent configuration is upgraded with the Decision model v5.1.9 and Name Matcher v1.1.6, which include:

  • Preprocessor to remove spelled-out numbers

  • Enhanced middle name matching with subset logic, for example, "John Paul Smith" vs. "John Paul Michael Smith" → score: 0.98

  • Refactored calculate_similarity() method for better readability and maintainability

  • Extended ORG preprocessing that removes mid-string text from configurable trailing legal-form starters, such as "spolka z"

Version 3.3.4

About Release

• Release date: July 3, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara v3.3.4 release aims to improve address matching flexibility, enhance entity recognition accuracy, enrich decision traceability, and support better precision in risk screening.

• The logic is updated to parse known SWIFT tag values inside FUF content, primarily affecting F tags, such as SWF_4_50F and SWF_4_59F. The full address is now correctly set after parsing the SWIFT content.

• To better handle * characters within the content, we updated the logic by removing 'New Lines From 36th Positions'.

• A detailed decision narrative is now included in the HTML report and hit decision response, making Tara's decision-making fully transparent without requiring inspection of underlying data objects. A new optional additionalInfo field is available in both the response object and the HTML report. You can also enable this feature in the configuration UI.

• Support is added for partial (fuzzy) address match scoring when external services return no results or are not configured. The enhancement helps escalate potentially suspicious transactions even when exact address matches are unavailable.

  • A probabilistic score is calculated using string similarity and structured data from the external service.

  • The addressMatch rule object now includes a new score parameter, and a sample custom rule is available for integration.

• To prevent typing errors and ensure valid, standardized country entries, the free-text input for the Blacklisted countries field is replaced with a pre-populated country selection with validation and automatic suggestions available.

• The narrative is updated for hits with empty tag content, improving clarity.

• During historical data upload, analyst comments from messages.csv can now be applied to all hits, not just records with a single hit. A new setting in the configuration JSON controls this behavior, which is enabled by default.

• The NER model is upgraded to v2.4.0 with the following improvements:

  • Enhanced model prediction accuracy for U.S. and Maldives addresses, Middle Eastern company names, and long personal names

  • Resolved BIC recognition issue where “ARQUITECTOS” was incorrectly classified as a BIC

  • Improved recognition quality when named tags are present in text, for example, "/ABIC/SBOSUS33 /NAME/Truist Financial MEXICO MEXICO"

  • Disabled merging of adjacent PER and ORG entities during post-processing

  • New post-processor to filter out overlapping entities

• The AI Agent configuration is upgraded with the Decision model v5.1.7, which includes:

  • Logic to stop parsing NER-based addresses when screened location or enrichment locations are already present, preventing incorrect country attribution. For example, "PHOENIX AZ" was misidentified as "Azerbaijan".

  • Improved Libpostal parsing of country names in SWIFT format where an ISO2 country code is followed by a slash and city name, for example, US/Boston.

• The Name Matcher is upgraded to v1.1.4, incorporating the following updates:

  • Resolved issue where share listing information was not excluded from ORG name comparisons

  • Revised list of stop words

  • Nickname support for Russian, Italian, and French names

  • Improved handling of company acronyms and name tokens split by spaces

Bug fixes

• Resolved the date conversion issue in the CLDS report by expanding the range of supported date formats.

• Fixed the issue in computing full addresses.

• Fixed the issue in entity resolution where the hit text displayed a country name, but the actual match was based on the ISO2/ISO3 code, leading to incorrect entity classification.

• Resolved the failure in saving large (> 64 characters) tag name values to the pss_alerts_analytics_v1 Data Store. Tag names are now truncated from the beginning, cleanly breaking on slashes.

• Fixed the compatibility issue where the Geocoder returned no Location object for unresolved addresses, causing Decision models v5.1.2 and earlier to fail. A blank Location object is now inserted to ensure backward compatibility.

• Resolved the entity resolution issue where the NER-detected entity did not directly match the hit text or sanctioned name, for example, "AL-RASHID" versus "ALRASHID" or "Al'RASHID".

Version 3.3.3

About Release

• Release date: May 30, 2025
• Compatible platform versions: Work.AI 10.2.8+

The release focuses on enhancing escalation capabilities based on analyst decisions, improving data processing accuracy, and upgrading models.

• The option to ignore middle names when comparing names is added to the AI Agent configutation, reducing false positives during entity matching.

• The tag content preprocessing is improved by removing the 36th-character newline before sending multi-tag input to the NER engine. The update addresses SWIFT line limitations and ensures accurate hit text alignment with modified tag content, enhancing entity recognition accuracy.

• To improve the error rate, we added the ability to automatically escalate hits. Now, hits are escalated if a hit hash was disabled by an operator for any reason or never released by an analyst. When configuring the AI Agent, you can now choose between auto-release, auto-escalate, or both options, ensuring no relevant hits are overlooked.

• To maintain audit trails, we enabled saving analyst comments from historical data upload and using them in decision review hit comments. The changes are as follows:

  • Comments are loaded from a new T_COMMENTS column of the hits.csv or messages.csv files.

  • A message-level analyst comment is only used if a message has exactly one hit.

  • A hit-level comment takes precedence over message-level comments.

  • For the same hit hash, only the last hit comment stored in the hash_comment column of the pss_clds_hash_analytics_v1 Data Store is used.

  • In the Output configuration section, you can now select using a custom continuous learning (CLDS) report template instead of generating a report with standard columns.

• The Tara API is updated to v1.2.5. No changes to input and output contracts are applied.

• The AI Agent configuration is upgraded with the Decision model v5.1.3 and Name Matcher v1.1.2 delivering the following improvements:

  • Adding cities to geo term preprocessing.

  • Introducing a specific preprocessor to remove currencies.

  • Revised preprocessing logic to improve overall accuracy in entity detection.

• The NER model is upgraded to v2.3.0 incorporating the following changes:

  • Improved model predictions for the Organization (ORG), Location (LOC), and Person (PER) types.

  • Improved recognition of Middle Eastern personal and company names.

  • Improved recognition of personal and company names in cases when input breaks the expected SWIFT field format.

Bug fixes:

• Fixed the configuration issue with quality control (QC) and CLDS reports when switching between data collection options. As a result, when consecutive reports are run, the date range configuration chosen in the first run does not supersede the configuration choice made in the second run.

• Resolved the issue with incorrect offsets after merging entities with IDs, enhancing the reliability of entity annotations.

• Resolved a number of issues by updating Name Matcher to v1.1.2:

  • Resolved the issue with the Geocoder producing no results when parsing enrichment data.

  • Resolved the issue with a country removed in one name and preserved in another.

  • Fixed the issue with the individual name's low score if matching on initials.

  • Fixed the issue with score calculation for names with uncommon tokens.

Version 3.3.2

About Release

• Release date: May 15, 2025
• Compatible platform versions: Work.AI 10.2.8+

The release aims at improving rule control, historical decision traceability, address matching logic, and decision-making processes. Mind that the release contains the updated default rule file, though you can continue using the rule file from Tara v3.3.1 (not earlier versions).

• To provide greater flexibility and precision in decision-making processes, the rule file now controls which decision score to use. A score is assigned in the decision object per rule by overriding RulesUtils.resolve and decision functions with a score parameter.

• We migrated the PSS QC Report, PSS CLDS Report, and PSS CLDS Ingest Historical Decisions Business Processes to a schema-based implementation. Now, all of Tara's Business Processes have contracts for improved data consistency and process performance.

• To increase the stability and accuracy of address matching in complex and incomplete data scenarios, the following improvements are introduced:

  • The address matcher rule is available now.

  • Exceptions occurring in specific scenarios when using the Geocoder option are resolved.

  • To enhance the address matching logic on the sanction side, the name field is used if the type is LOCATION and no address field is provided.

  • For address matching using location objects, a full address is built using the city, state, and country fields.

• We updated the data upload archive path to include the processed date. You can now locate historical data more easily and quickly.

• The AI Agent configuration is upgraded with the Decision model v5.1.2. It removes unused code related to named entity recognition and hit text calculation and eliminates obsolete raw output features from the model response.

• The HTML report header is updated for a more consistent and user-friendly presentation of the reported data.

Version 3.3.1

About Release

• Release date: May 1, 2025
• Compatible platform versions: Work.AI 10.2.8+

The release provides compatibility with RHEL 7.9, streamlines rule management, and improves reporting and analytics capabilities.

• Tara now fully supports RHEL 7.9, with previous Python library issues addressed to provide a smoother installation and runtime experience.

• To simplify the processing pipeline, we removed the filtering step from the PSS core Business Process, along with all associated filter processing, utilities, and configuration. You can leverage the same filtering functionality by writing rules in your custom rule file.

• The default rule file is updated to reflect customer feedback, ensuring alignment with actual business needs and providing a more tailored configuration out of the box.

• The parser functionality is enhanced to bypass execution of model steps when tag content is empty:

  • The parser now makes an in-place decision for hits with empty tag content, while other hits continue through the standard processing path.

  • The parser bot contract is updated to reflect changes, and the analytics bot contract now outputs a final message object.

• To support auditability and compliance monitoring, we introduced the hash_update_date column in the continuous learning (CLDS) analytics Data Store and CLDS reports.

  • When historical-decisions/hash_status_change_file.csv is uploaded, hash_update_date is populated instead of hash_enabled_date. The latter now only records timestamps when a hash is enabled through Tara's real-time processing.

  • The is_auto_disabled column is no longer set but remains present in the pss_clds_hash_analytics_v1 Data Store for backward compatibility.

• Unused code and classes related to Fircosoft data processing are removed, reducing maintenance overhead.

• To offer improved visibility and more comprehensive analytics for compliance teams, CLDS report enhancements include:

  • Correct display of the minimum counter value

  • Updated CLDS template file that now contains an updated alias for alert_operator and a new hash_all_released column

• The logic for processing maximum hits is updated to ensure data accuracy:

  • Validation is moved from the parsing step to the input processing step.

  • Requests exceeding the maximum number of hits are now rejected and excluded from further processing and HTML reports.

  • All invalid requests are logged in the request processing table with the FAILED status.

• The NER model is upgraded to v2.2.0, featuring the following changes:

  • Enhanced predictions for entity types ORG, LOC, and PER, including:

    • Improved recognition of "DMCC" and similar legal endings as company suffixes

    • More accurate location extraction in case of repetitions, such as "Tunis, Tunis, TU, Tunisia"

    • Better handling of complex personal names, such as "Ahmed Fuad Bin Sheikh" or "Ricardo S. De La Cuba"

  • Resolved the issue where the NER model was stuck when encountering excessively long tokens in input.

• The AI Agent configuration is upgraded with the Decision model v5.1.1 and Name Matcher v1.1.1, delivering the following improvements:

  • Introduced more robust handling of middle names with parameter-based control over individual name comparison behavior.

  • Fixed the issue in the base similarity calculation that caused incorrect comparisons for Arabic names starting with "al-".

  • Disabled application of the rare token overlap logic for concatenated names with ambiguous stopwords.

  • Updated the logic for handling RARE or SCREENED token overlaps: name matches are now returned if two or more tokens match, regardless of their frequency.

  • Improved preprocessing for organization names: the text is now removed after legal forms if found at the end of a name.

  • Added support for common word variations, such as "NATL == NATIONS == NATIONAL".

  • Removed the functionality related to using the external Babel name matcher.

  • Updated the Python environment to python3.11.5-pss-v4 and replaced the C++ Double Metaphone library with a pure Python alternative to ensure compatibility with RHEL 7.9.

Bug fixes:

• Resolved the issue when invalid data could be saved in the threshold fields on the Model tab. Field validation now correctly prevents invalid entries, ensuring data integrity.

• Resolved the quality control (QC) report issues with wrong values in the hit_name_matcher_score and hit_type_confidence_score columns to provide more accurate quality control metrics.

• Fixed issues in CLDS and QC reports that caused failures when running against requests generated with Tara versions prior to v3.3.0.

• Fixed the hash mismatch issue caused by tag content ending with a backslash before a newline character. Newline conversion is moved from the generic message converter to the SWIFT message parser to ensure consistent hash calculation.

Version 3.3.0

About Release

• Release date: April 4, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara v3.3.0 release extends the address verification capabilities and introduces a way to set custom disposition rules. Mind that the release is compatible with models of v5.0.2 or higher versions, while older models will no longer work with the solution.

• Tara now supports multiple locations for screened and sanction data, entailing the following changes:

  • Tara's input contracts were updated, and its REST API can now handle multiple addresses in both screened and sanctioned data for each alert. For the purpose, two new fields were introduced: screenedData.locations and sanctionData.locations. The PSS Business Process, however, remains backward compatible, and older inputs will not fail.

  • Geocoder now searches all screened addresses and saves multiple search results. The isLocationValid parameter has been split into isCountryValid to define whether the found country can be used for location match or mismatch and isAddressValid to confirm whether a street address is a real postal address for a building.

  • The v5.0.2 Decision model now supports the mentioned new location fields and geocoder search results. The v5.0.2 model is backward compatible with Business Processes prior to Tara v3.3.0 and works with the older Geocoder version. However, the v5.0.1 and earlier models cannot recognize the new location fields added in the PSS Business Process for Tara v.3.3.0.

  • The rule mapping was updated to reflect supported multiple locations, while unused fields were removed.

• The new address comparison capability allows you to verify if the screened and sanctioned addresses match using one of the address matchers: Geocoder API, in-house address verification API, or literal string comparison. You can configure the capability and choose an address matcher on the Model step in Tara's configuration user interface.

  To support the address comparison feature, the following updates were made to Tara's rule configuration: the locationMatch object was renamed into countryMatch, and the new addressMatch object was added.

• To allow you to define custom disposition rules, we provided a possibility to set custom attributes in Tara's input. For that, new customAttributes maps were added to JSON requests and made available in the rules.

Bug fixes:

• Fixed the issue with the last token of addresses incorrectly identified as an ISO country code (for instance, in the 123 AMSTERDAM 1077 BL address string, the correct country is NL, not BL).

Version 3.2.8

About Release

• Release date: March 28, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara v3.2.8 patch aims to improve data normalization, models' accuracy, and overall stability to enhance screening reliability and reduce false positives.

• In CLDS, the ! character is now excluded from hit hash calculation, addressing cases where the | delimiter was converted into a ! symbol. To ensure consistent hashing, the ! character is added to the list of special characters normalized before hash generation.

• The AI Agent configuration is upgraded with the Decision model v5.0.3, incorporating:

  • Support for multiple locations in the new screenedData, sanctionData, and enrichmentData fields, now available in the Business Process schema.

  • An updated Name Matcher v1.0.5 that fixes the incorrect removal of legal endings, resolves a potential execution issue, and revises the stop words list.

• The AI Agent configuration is upgraded with the NER model v2.1.8, featuring the following changes:

  • Handling financial code abbreviations, covering account and bank identifiers, reference numbers, amount and currency details, compliance and regulatory markers, and more.

  • Addressing the issue where incorrect PER predictions occurred when a name was concatenated with numeric strings.

  • Introducing post-processing for the 35B SWIFT field and resolving the issue where "ISIN" and similar markers were incorrectly predicted as ORG.

Bug fixes:

• Resolved the issue with incorrect FUF content parsing when the content included a \r character inside a tag's value. The fix ensures proper parsing of FUF content, improving data quality and preventing potential downstream processing errors.

Version 3.2.7

About Release

• Release date: March 19, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara v3.2.7 release focuses on enhancing data integrity, auditability, and operational efficiency in CLDS, improving decision-making processes, and refining entity recognition capabilities.

• As part of CLDS enhancements, we added the ability to upload and process a hash status feedback file to improve the management of historical hash decisions.

  • The PSS CLDS Ingest Historical Decisions Business Process now checks for an additional hash_status_change_{date}.csv file each time it runs. If the file is provided, it is fetched together with historical files and processed.

  • Support for multiple file uploads allows simultaneous submission of multiple feedback files.

  • The pss_clds_hash_analytics_v1 Data Store includes new columns to record the updated status, timestamp, and operator who requested the change, improving auditability.

  • The pss_dr_analytics_v1 Data Store now allows recording actions. Failure events are also logged, providing better visibility into hash status changes.

  • CLDS reports now include the hash_operator column to provide insight into status change requests.

  • Hash consistency for hit_type = "national id" is improved to ensure that hashes remain identical regardless of spaces in the input, preventing discrepancies in identity matching.

• Configurable analyst exclusion is now available in continuous learning decision calculations. In the Reapplication Data section of the AI Agent configuration, you can set up analyst or operator names for exclusion from the decision-making process when determining whether a hit qualifies for reapplication.

• Support for the NTN ID NER ID type is introduced, addressing the issue where Tara could not process NTN IDs from input data.

• The decision object now displays the type confidence score instead of the name matcher score for type mismatch rules, ensuring more precise scoring.

• The AI Agent configuration is upgraded with the NER model v2.1.5, incorporating the following changes:

  • Improved ORG postprocessor logic prevents country codes (for example, AE or AG) from being incorrectly classified as business entities.

  • Handling of CUSIP numbers is added, introducing the ID_CUSIP label to apply dispositioning logic based on ID information.

  • Recognition of INN numbers is now supported with the help of the ID_INN label.

  • Legal Entity Identifier (LEI) is now handled as a special ID label, enabling identification of legal entities participating in financial transactions.

  • BIC code detection is enhanced to recognize codes preceded by the "B" character, improving Tara's ability to process BICs.

  • Entity-level precision, recall, and F1-score metrics are now used to evaluate NER models, ensuring more accurate assessment by considering entire entity spans rather than individual token predictions.

Bug fixes:

• Resolved country-related issues by updating the internal location lookup logic for pipe-separated countries. Additionally, the conversion logic for the special region code "UA-43" is corrected to prevent errors related to the Crimea region.

Version 3.2.6

About Release

• Release date: February 28, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.2.6 release refines entity recognition and improves logging and reporting capabilities. Key changes include enhancements to hit entity calculations, updated models, and corrected issues in decision-making and reporting mechanisms.

• To enhance data accuracy and processing precision, a retained hit split on ID is introduced. A hit split now occurs only if the hit text is fully inside the ID.

• Improved error logging for CLDS historical data uploads now enables email alerts with full exception details, improving issue identification and resolution.

• To enhance traceability, the Operator column is added to the CLDS report feedback file for each cleared hit.

• The hit entity calculation is updated to reduce the number of splits per hit, increasing the automation rate.

  • Hits are no longer split if the hit text or watchlist entity name exactly matches a returned NER entity.

  • Overlaps between hit text and watchlist entity names are now used as significant tokens to determine hit entities. Note that fewer splits might result in a higher automation rate.

• The AI Agent configuration is upgraded with the Decision model v5.0.1 incorporating Name Matcher v1.0.3, which includes:

  • Fixes for concatenated score computation—no penalty is applied when two names match perfectly, for example, "Trust Bank" versus "Trustbank".

  • Updated lists of legal forms and stop words.

  • A fix for incorrect matches caused by initials, for example, "Kennedy" versus "J K Rowling" now correctly results in a mismatch.

  • Adjusted preprocessing to prevent the removal of company stop words from personal names.

  • A new check for complete name matches before preprocessing begins.

• The NER model is upgraded to v2.1.2 improving recognition of national IDs, National Tax Numbers (NTN), and invoice numbers. Additionally, 19 new ID labels are added as special tokens, enabling Tara to recognize and disposition hits on these values more effectively.

Bug fixes:

• Resolved the issue where an incorrect name matcher score was used when parsing the model output. The highest score is now selected from model results regardless of the data source, though this change might slightly impact the automation rate.

• Addressed a potential exception in CLDS uploads caused by incorrect file extensions.

• Fixed the issue in collecting the type confidence score when the screened type is NAME.

• Resolved the issue with missing decision values for split hits. Now, the hit_decision, hit_decision_code, hit_decision_comment, and hit_name_matcher_score columns now correctly display values for split hits in QC reports.

Version 3.2.5

About Release

• Release date: February 10, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.2.5 release includes a reworked analytics dashboard, better report handling, optimized storage organization, and upgrades to the decision and NER models.

• To improve analysis and accelerate the decision audit process, we updated the PSS analytics dashboard.

  • A new Superset analytics dashboard is introduced with more detailed adjudication information. The dashboard contains filters for various article lists, countries, and search types, detailed reports for diving into specific data points, and charts and graphs visualizing screening results.

  • The request_insert_time, status, and tara_version columns are added to the Message Analytics Data Store (ds_pss_messages_analytics_v1), with updated analytics table indexes to improve query performance. The changes streamline data retrieval and reduce query complexity, resulting in a faster and more efficient dashboard.

• All Tara-related S3 files are moved under a single root directory: doc-upload/payment_sanctions_screening. The previous doc-upload/sanctions directory is no longer used. Updated locations for S3 files follow the new structure:

  • HTML reports: doc-upload/payment_sanctions_screening/reports/html

  • Rules: doc-upload/payment_sanctions_screening/rules and doc-upload/payment_sanctions_screening/{$version}/rules

  • CLDS historical upload: doc-upload/payment_sanctions_screening/historical-decisions

  • QC and CLDS templates: doc-upload/payment_sanctions_screening/reports/templates

  • QC reports: doc-upload/payment_sanctions_screening/reports/qc

  • CLDS reports: doc-upload/payment_sanctions_screening/reports/clds

• Previous QC and CLDS reports are now automatically archived before new ones are generated. The PSS CLDS Report and QC Report Business Processes are updated to move previously generated reports to the reports_archive directory in S3.

• To enhance data security and restrict client access to other MinIO storage, we now provide a custom S3 bucket for HTML, QC, and CLDS reports, as well as CLDS historical uploads. Rules and report template locations remain unchanged, and the sub-directory structure within a custom bucket mirrors that of the doc-upload bucket.

• The AI Agent configuration is upgraded with the NER model v2.1.0 featuring:

  • Improved recognition of multiple people's names appearing consecutively

  • Better differentiation between people and organizations

  • Enhanced detection of common phrases that do not contain entities

• The Decision model is updated to v5.0.0 to remove obsolete features and UI configurations. All required configurations are migrated to the core Payment Sanction Screening Business Process UI.

Bug fixes:

• Fixed exceptions related to broken content responses and missing message types.

• Fixed CLDS report issues:

  • Resolved missing MessageText values in the CLDS report when tag content was not parsed. Now, when a hit is good for reapplication, tag content is correctly parsed and included.

  • Addressed missing MinimumCounter values when the hash status was NOT_SET. If the MinimumCounter or MinimumAnalyst values are missing or set to 0, they are now added from the configuration to the cldsInfo object.

  • CLDS resolved hit comments now display the number of times a hit was dispositioned and the number of analysts involved.

  • Date formats are synchronized for Enabled date and Created date in CLDS reports. The last_timestamp field in QC reports is set to the yyyy-MM-ddTHH:mm:ss.SSS format.

• Fixed the issue where a hit on the abbreviated name "C. I. D. E. I. M" was not identified correctly. The hit text calculation logic was updated with the SignificantOverlap function, ensuring proper token order comparison between hit text and NER text.

Version 3.2.4

About Release

• Release date: January 28, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.2.4 release brings key enhancements to security, data handling, and overall system stability. Highlights include improved encryption for sensitive data and centralized model configuration.

• Blacklisted keywords and country configurations are moved from the Decision model to the Business Process level. You can set up blacklisted entities and keywords in the Model section of the AI Agent configuration. Support for blacklisted companies is removed, and the blacklisted company rule is no longer available. To prevent failures, you might need to update your custom rule file manually.

• Column-level encryption for Personally Identifiable Information (PII) data is introduced for specific database columns. A new annotation marks fields requiring encryption, allowing seamless integration with the correct JDBC connection settings.

• The synthetic data generator is refined, with new scenarios added and comments updated for improved readability.

• The source system ID input message field is included in the JSON input request contract, and a corresponding column is added to the CLDS report.

• The Decision model is updated to v4.4.2 to remove obsolete features that previously caused errors with specific input data.

• The rule object now contains new WLE fields (SourceEntryId, sourceKeywords, and source) for better data handling.

• The AI Agent configuration is upgraded with the NER model v2.0.7, featuring the improved SWIFT tag post-processor for better entity recognition.

Bug fixes:

• Resolved the issue with "Hong Kong" recognized as "CN (China)" instead of "HK".

• Corrected alert comments in cases of invalid input flow and successful decision reapplication flow.

• Resolved the issue where the Decision model returned an unexpected error and displayed "Hit was not processed", leaving the Name Matcher section of the rule object empty.

Version 3.2.3

About Release

• Release date: January 17, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara v3.2.3 patch introduces vital improvements for Continuous Learning Data Store (CLDS) processes, alert and report enhancements and addresses previously reported issues.

• The CLDS process was adjusted to reduce the processing time and improve the feature performance, including the following changes:

  • A new PSS CLDS report Business Process was introduced, enabling the generation of CLDS reports on a scheduled basis. The report generation settings are configurable in the AI Agent interface. You can now run reports on demand or on a defined schedule. Generated reports include predefined columns and display only matches from historical hash data.

  • The CLDS historical upload process was reengineered to ensure CLDS execution completes within five seconds. To support the update, a new CLDS analytics Data Store was implemented to store historical upload statistics. All data points required for real-time CLDS processing are now calculated and stored in the analytics Data Store.

  • Real-time CLDS processing was rebuilt to remove real-time searches against historical data and replace the rule engine with simplified predefined conditions. The configuration form was updated to reflect the revised workflow. Real-time requests now query the new CLDS analytics Data Store. Updates were also applied to the Data Stores to enhance CLDS testing capabilities, as well as data and decision analysis.

• Decision-based alert comments are now stored in the Data Store, which improves traceability and simplifies their analysis.

• The AI Agent configuration is upgraded with the NER model v2.0.6, featuring improvements to the hit text and entitity matching capabilities to enhance the overall model accuracy and reliability. The adjustments are as follows:

  • Exact matches are now allowed for two-letter hit text when the type is ID, such as "SB" in "1234567SBK89".

  • The default hit text length for exact sub-token matches is reduced from 4 to 3 characters.

  • Single-letter tokens are excluded from similarity score calculations, for example, "SBK" versus "B".

  • For improved NER entity type detection, the NER model input excludes SWIFT markers, such as "1/" and "2/", from the tag content.

Bug fixes:

• Resolved an issue with the post-processor that caused duplicate location text.

Version 3.2.2

About Release

• Release date: January 2, 2025
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.2.2 patch enhances the accuracy of Tara's decision-making and improves configuration handling, reducing error rates and ensuring compliance with international standards.

• The internal country parser now identifies countries using only ISO2 and ISO3 codes, preventing issues when processing countries with multiple tokens in their names.

• The Decision model is upgraded to v4.4.1, incorporating Name Matcher v1.0.2.

  • The updates address the issue in concatenated score computation, reducing misidentification risks by ensuring no penalty is applied to names that match perfectly, such as "Trust Bank" and "Trustbank".

  • The list of legal forms is updated to align with the latest standards.

• The AI Agent configuration is upgraded with the NER model v2.0.5, which features increased confidence thresholds and a post-processor specifically designed for Danish addresses resembling Slavic last names.

• The 53B SWIFT tag dictionary is updated to address incorrect decision-making issues.

Bug fixes:

• Resolved the issue with Tara not splitting hits. The multiple hit split processing logic is updated to include token breaks by new lines.

• Resolved the issue with Tara failing to parse the 59F tag, resulting in incorrect decisions. The fixes include:

  • Stripping the SWIFT marker from hit text during parsing

  • Improved country validation, returning an empty string for invalid lookups

  • Enhanced city and state lookup, returning true if any data is found

Version 3.2.1

About Release

• Release date: December 5, 2024
• Compatible platform versions: Work.AI 10.2.8+

The Tara 3.2.1 patch improves rule management, data enrichment, location handling, and decision-making capabilities, while addressing key issues to ensure smoother and more transparent operations.

• To enhance the ability to create robust and reliable NONENTITY handling rules, confidence score handling was moved from the NER results processing Business Process step to the rule file. The default rule file (rules_v1.drl) was updated to incorporate type-matching confidence scores directly within the rules. The configuration option for setting the NER type confidence threshold was removed from the UI.

• To improve traceability and accuracy in quality control reporting, the quality control (QC) report was modified to include filtered out hits from decision reapplication or filtering rules. The update also resolves the issue of mismatched hit counts between decision or gold decision objects and message objects caused by hits removed during decision reapplication or filtering.

• For workflow optimization, the Duplicate response step was removed from the core Payment Sanction Screening Business Process due to the adoption of the schema-based implementation. The Send response step is now placed directly before the Save analytics step.

• To address limited location information and ensure better screening results even with incomplete inputs, enhancements were made to the sanctioned and screened country lookup, including:

  • City and state or province lookup, enabling country derivation from cities or states when country data is missing, for example, city "Abali" > Iran, country code "IR". Additionally, city dictionaries were added with data available in multiple formats, such as JSON and CSV.

  • Validating screened city values when parsing data and deriving possible countries when country data is invalid or missing. A sanctioned location can only have one country, but a screened location can have multiple possible countries if the same location name exists in multiple countries, for example, city "London" > United Kingdom, United States, or Canada, with country codes "GB|US|CA".

  • Identifying country codes embedded in street addresses, for example, "2213, PORT ABACUS, POCHA, 4211 LY" > Libya, country code "LY".

• The Validate every address (even if Country is already known)? option to enable the Geocoder to always run was added to the Data Enrichment configuration screen. The feature enhances data validation flexibility and geocoding accuracy.

• Alert-level decision comments with summaries of hit decisions are now available as part of returned decisions from Tara, providing comprehensive information for easier review.

• The multiple hit split logic was updated, improving accuracy in handling complex hit scenarios.

  • The logic for creating split hits is now more flexible, allowing splits on nonentities and even if a single token in hit text exactly matches that of an entity.

  • No exact match tokens in hit text are discarded anymore.

  Note that the update increases the number of splits and might potentially reduce the automation rate.

• The AI Agent configuration was upgraded with the NER model v2.0.3, resolving the issues related to type mismatches and NONENTITY being incorrectly returned as an address.

Bug fixes:

• Resolved the issue where an alert with two hits displayed two split parts of one hit but omitted the second hit in the HTML report. Unused fields were also removed from the report template, ensuring a complete and accurate representation of alert data.

Version 3.2.0

About Release

• Release date: November 15, 2024
• Compatible platform versions: Work.AI 10.2.8+

info

Tara v3.2 is built on top of v3.1.21.

The 3.2.0 release strengthens Tara's capabilities and ensures compliance with modern security standards, delivering a more robust and efficient workflow enhanced with a contract-based approach and upgraded models. Additional improvements focus on increasing system reliability, data interoperability, and security and reducing operational complexity.

• The Payment Sanctions Screening workflow is redesigned to adopt contract-based processing, ensuring better maintainability.

  • Contracts are introduced for each step of the core Business Process.

  • A schema from the Sanctions API client is integrated to enable seamless data exchange.

  • To improve data handling, contract-based steps now use JSON-based input and output data, replacing legacy Data Store operations.

• The AI Agent configuration is upgraded with the NER model v2.0.1 featuring:

  • An updated Python installation and dependencies that address security and licensing vulnerabilities

  • Modified prediction logic enhancing entity identification accuracy

  • Post-processor improvements for combining consecutive entities

• To enhance Tara's ability to identify and process sanctioned entities, the Decision model is updated to v4.4.0, which includes:

  • A Python installation with the latest improvements.

  • Fixes for license and security issues identified by Snyk.

  • Name Matcher v1.0.1, now supporting expanded variations of legal form names. For example, "società a responsibilità limitata" matches SRL and its variants.

Version 3.1.21

About Release

• Release date: November 11, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.21 patch focuses on improving data processing efficiency, streamlining rule management, and strengthening system resilience. The enhancements to Tara's models result in faster, more accurate processing with reduced downtime, ensuring smoother operations and more consistent data-driven decision-making.

• To eliminate redundant processing, we removed unused logic for Additional Data and Account Holder Name from the Business Process workflow. The Decision model is also upgraded to version 4.3.8, which omits the usage of Additional and Self-Service data along with Account Holder data.

• The rule management mechanism was enhanced by introducing version control for the custom rule file.

  • The default rule file is now stored in a versioned directory, such as doc-upload/sanctions/{$tara_version}/rules/rules_v1.drl. Modifying the default file is not recommended.

  • You can place a custom rule file in a designated directory for it not to be overwritten during updates. This custom file will take precedence over the default one.

  For more details, see Install and configure | Model and Make decisions.

• The AI Agent configuration is upgraded with the NER model v1.6.17 that addresses the ID recognition issue caused by the ID post-processor’s reliance on specific SWIFT tag names. The updated model enhances the accuracy of digital identity verification and ensures consistent data extraction.

• We reverted a change introduced in Tara v3.1.19 that previously batched historical data queries. While reducing the memory consumption, the batching also led to significantly slower CLDS execution. Instead, we increased the heap memory size from 512 MB to 1024 MB in the configuration YML file to prevent rare out-of-memory exceptions. Additionally, the select query for fetching historical data is updated to load all results in one query, improving execution speed and efficiency.

Bug fixes:

• Resolved the 503 error when accessing the default rule file from S3 due to connectivity issues with S3 MinIO. If S3 is temporarily unavailable, the system loads the latest known rule file from cache to ensure uninterrupted access.

• Fixed the issue when the connector failed to return a response in case the input originated from the connector. Now, when Tara receives input from a connector, it searches for JSON input within the connector_request field. The output is then generated in the connector_response column and sent back to the connector.

Version 3.1.20

About Release

• Release date: October 30, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.20 release introduces enhancements to streamline data processing, improve parsing accuracy, and reduce system load for greater efficiency. The updates target the reapplication process, hit filtering, data parsing, and country code conversion.

• To manage data processing more efficiently, we added the ability to compare the SUBTYPE field from historical maps with the message.subtype parameter of incoming requests during the reapplication process. To configure the setting, go to the Reapplication Data configuration screen and specify SUBTYPE in the Message tags to compare section.

• The filtering mechanism is enhanced to bypass further processing if all hits are filtered out. When no unfiltered hits remain, the message is directed immediately to the Generate report or Apply rules steps. This improvement helps save processing time by avoiding unnecessary steps.

• To comply with international standards, the model now supports converting country values to ISO2 codes. To take advantage of the feature, make sure the rules object contains countries in the correct format.

• To ensure the user-defined address data remains unmodified and accurate, the parser is set not to overwrite user-defined addresses.

• Multiple improvements are made to the ISO content parser to address issues in parsing-related data:

  • Resolved issues with extracting values from unrelated parent tags and retrieving nested parent tags.

  • Updated the FUF dictionary to capitalize attribute names as needed, ensuring that terms like PAYMENT_TRANSFER are recognized accurately.

Bug fixes:

• Fixed the exception in the QCReport function that occurred when no custom report template was provided and query data returned empty results. The fix ensures that QCReport handles empty data gracefully, preventing unexpected errors.

Version 3.1.19

About Release

• Release date: October 21, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.19 release focuses on analytics, search efficiency, decision-making precision improvements and general system optimization.

• To enhance Tara's analytics, we added tag_name and content_format columns to pss_alert_analytics Data Store.

• To prevent out-of-memory exceptions when searching over long periods, the batch size is made configurable, with the default value of 100. When configuring the Reapplication Data settings, you can set the maximum number of records returned from a search query in the Batch size field.

• The configuration is upgraded with the Decision model v4.3.7, which includes Name Matcher updated to v0.5.6. Stop words for individuals and the term mapping list are revised to deliver more accurate decisions.

Bug fixes:

• Fixed the issue with the missing additionalInfo field when converting from a flat format file to a JSON request.

• Removed unused repositories and services and cleaned up LiquiBase YAML files.

Version 3.1.18

About Release

• Release date: October 16, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.18 release aims to enhance execution of Java-based tasks by migrating Tara processors to the Java Native Worker (JNW) solution. For more details on JNW, see the documentation. For instructions on migrating an existing project based on the ODF 2 framework, see ODF 2 integration with Java Native Worker.

For Tara v3.1.18, you cannot use rules from previous versions. You also must update import statements.

Version 3.1.17

About Release

• Release date: October 11, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.17 release ensures more accurate name matching and better handling of edge cases, offering higher precision in decision-making processes.

To improve matching accuracy, we upgraded the AI Agent configuration with the Decision model v4.3.5, containing Name Matcher updated to v0.5.4 with the following changes:

• Name Matcher no longer applies phonetic similarity for tokens shorter than three letters and does not treat short names as potential acronyms.

• The acronym comparison logic is improved by excluding legal forms from comparison. For example, DIL LLC versus DREAMWAY INDUSTRIES LIMITED is now considered a mismatch.

• A penalty is now applied for incorrect word order in names. For example, RESI versus Environmental Systems Research Institute is recognized as a mismatch.

Bug fixes:

• Fixed the issue with incorrect names used in decision-making. For multipart hits, filtering hits of NONENTITY types now stops when other relevant types are present.

Version 3.1.16

About Release

• Release date: October 2, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The 3.1.16 release introduces support for using Tara in the AI Agent chat-based interface, provides improved report generation, and contains the upgraded Natural Entity Recognition (NER) model.

• The core Business Process workflow is updated.

  The HTML report generation and analytics/billing are split into separate steps, improving workflow management.

• Including an HTML report link in the REST API response JSON allows viewing reports directly from the API response. To try a new feature, select Include HTML Report link in API response during the output configuration.

• The AI Agent configuration is upgraded with the NER model v1.6.15 for better performance.

Version 3.1.15

About Release

• Release date: September 30, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara 3.1.15 patch aims to optimize decision-making processes and provide more accurate scoring mechanisms.

• The ability to configure a historical resolve decision type in the CLDS flow enables you to define custom decision logic that aligns with your specific business needs. When configuring the Reapplication Data settings, set historical decision types that will be treated as resolved by an analyst.

• The AI Agent configuration is upgraded with the Decision model v4.3.4 containing Name Matcher updated to v0.5.2. The updates provide revised scores for cases when the match_if_token_overlap setting is marked as screened. Now, if all tokens from the screened name are present in the WLE name the score will be greater than or equal to 0.95. Also, unused resources were removed to optimize the model performance.

Version 3.1.14

About Release

• Release date: September 26, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara 3.1.14 patch introduces several enhancements focusing on Quality Control (QC) reporting and improved usability. The updates make it easier to customize and manage QC reports according to specific needs:

• A new Business Process for generating a QC report is now integrated into Tara's existing workflows.

• The QC report settings are available during the output configuration in the UI:

  • The option to generate a report with standard columns allows you to create QC reports using a set of standard columns provided in a template CSV file. You can customize the report by modifying column names in the alias column, which is blank by default.

  Additionally, you can upload a custom CSV file based on the original template file, specifying only the columns required in the report.

  • The option to manage data collection provides a choice on how much data will be loaded in the report. You can select between all completed data for the current day (Today), the specified number of days (Last x days) or months (Last x months), or define a specific date range using the calendar widget (Custom). The Business process execution IDs option to input one or more Business Process run UUIDS allows generating results for those specific runs only.

  • The Maximum file size option enables you to define the maximum file size for generated reports. A report that exceeds 100 MB is automatically split into multiple CSV files.

• The unused pss_flat_results_v1 Data Store and its associated classes are deleted to optimize system performance and eliminate unnecessary data storage.

Bug fixes:

• Resolved the split-hit issue with multiple hit IDs in the tag content. The multi-hit split was not functioning correctly, and only the first ID was considered as a hit. Now, all relevant IDs are correctly recognized and processed.

Version 3.1.13

About Release

• Release date: September 25, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara 3.1.13 patch introduces a number of fixes to improve its models and matching logic:

• To make sure Tara uses the NER model output correctly, the hit text is no longer calculated when the NER model returns only an entity. The AI Agent considers an entity to be a hit irrespective of whether the hit text is a match or not.

• Tara's Decision model was updated to v4.3.3 to achieve better accuracy in processing the location-related data.

• In hit objects, the tagContent value of user-defined tags is no longer replaced with parsed values, making Tara's NER model more accurate in recognizing screened entities.

• To refine the matching logic, changes were made to how the NAME entity type is handled in the type_mismatch_wle_vessel rule. Now, if the input type is NAME and the WLE type is ORG/IND/LOC, the type match is true. Thus, Tara can attribute the entity type correctly as INDVIDUAL, ORAGNIZATION, LOCATION or as NAME.

Version 3.1.12

About Release

• Release date: September 17, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara 3.1.12 release focuses on improving decision-making, reporting, and matching capabilities and offers more flexibility in rule-based configurations.

• More precise NER parsing and hit text calculation for multipart tags result in better identification and categorization of entities, leading to more accurate and informed decisions.

• The option to generate a one-page report file allows to easily review all sanction hits for each sanction alert in a single report. To utilize new functionality, while configuring Tara, select the One-page report with all Hits report type in the Output settings.

• The option to extract country codes from IDs for location matching improves the accuracy of location-based operations, reducing errors and increasing automation efficiency. The new functionality is available in the Model settings during the AI Agent configuration.

• The AI Agent configuration was upgraded with the Decision model v4.3.2 containing Name Matcher updated to v0.5.0. The updates provide more advanced logic for handling token overlaps during name matching, increasing accuracy in scenarios where names contain similar or overlapping tokens. The model also supports multiple pipe-separated screened countries, for example, US|GB.

• To improve the granularity and accuracy of country-related data, multiple pipe-separated countries identified by the parser are now appended to the screened country field instead of overwriting, while multiple countries detected during the NER processes are included in the NER country data.

• To make writing rules easier, the NER confidence score was added to the type match object in the Rules Model JSON.

• The rules utility function checks the number of tokens, which enhances the flexibility and precision of the rule-based logic.

Version 3.1.8

About Release

• Release date: August 27, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara v3.1.8 release aims to increase the AI Agent's output quality:

• To refine the name matcher output and eliminate incorrect hit decisions, a comparison of legal forms was introduced for screened entities. The comparison results are available in the new LegalEndingsAreDifferent parameter added to the Rules object.

• The AI Agent configuration was upgraded with the classification model v4.3.0 and NER model v1.6.15. The updated classification model supports the introduction of the LegalEndingsAreDifferent parameter.

Bug fixes:

• Fixed the NullPointer exception at null hit.getTagName() that caused Java Native Worker to fail on the parsing step.

• Fixed the issues with split duplicates removed across multiple hits as introduced in 3.1.7 to make sure no hit decisions are missed.

Version 3.1.7

About Release

• Release date: August 9, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara v3.1.7 release aims to improve Tara's narrative, bringing it in line with the auditing and regulators' standards:

• In the screening results, the Entity marker is replaced with the Hit one, making the report consistent with the terminology generally accepted for screening system narratives.

• To avoid losing essential narrative data because of truncation and eliminate unnecessary duplicates, Tara no longer returns multiple hits if sub hits have identical names.

Version 3.1.0

About Release

• Release date: July 2, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

The Tara v3.1.0 release aims at enhancing efficiency, reliability, and overall performance and provides the following capabilities:

• Enhancements to the Decision Reapplication process allow for more precise and granular control, enabling partial message resolutions and more detailed analytics tracking. These changes are particularly beneficial for users dealing with complex payment messages and diverse data formats, such as SWIFT and SWIFT_MX.

• Model updates focus on improving the Named Entity Recognition (NER) and Decision models, enhancing their ability to accurately identify and process relevant entities. Due to refining preprocessing and post-processing steps and updating algorithms and matchers, the models are more robust and reliable, producing even more accurate results and increasing the solution's overall effectiveness.

• The updated ISO20022 parser enables more flexible tag name lookups, ensuring that Tara can seamlessly handle a wider variety of data formats.

• In addition to new features, several issues related to country matching, name concatenation, and entity detection have been resolved to ensure a smoother user experience.

New features

Business Processes:

• Updated the ISO20022 parser tag name lookup to enable finding ISO tag names regardless of the case used.

• Updated the AI Agent configuration > the Decision Reapplication step.

  • Added the Require tag content match for decision reapplication option (disabled by default) that requires the text of the payment message tag (for example, SWF_4_50K) to match the historical message for a decision to be reapplied.

  • Added the Decision Level option used to specify the level on which the decision reapplication will be applied—a hit level or a message one (by default, set to Hit Level).

  • Added a possibility to configure message tags for multiple formats.

  • Enabled ID-prefix matching.

• Updated the hit text logic to use a WLE name if the hit text contains a WLE name.

• Removed the redundant Read Settings bot step. All configurations are now read directly from a system-provided configuration JSON object, saving execution time.

• Updated the Named Entity Recognition (NER) model to v1.6.13.

• Updated the Decision model to v4.1.7.

Decision Reapplication

• Applied resolutions to individual hits instead of entire messages, allowing partial resolution of messages by Decision Reapplication.

• Added the decision_reapplied column in the Analytics table to record which hits and entire messages are resolved by Decision Reapplication.

• Added the support for the SWIFT format.

• Added the support for the SWIFT_NPP format.

• Added the support for the SWIFT_MX format:

  • Updated the ISO (XML) parser to better match tag content using the combination of the tag_name and hit_text values.

  • Updated the type dictionary to group related entity information together (for example, name, address, ID).

  • Updated rules to treat the Country type the same as Location.

• Introduced a solution to multiple hit-text logic. Hits are now split when multiple hit texts are found in the tag content, and each sub-hit is adjudicated separately. Sub-hits are merged back during decision processing. A NO_DECISION hit is returned if any sub-hit is NO_DECISION. The change may decrease the automation rate but prevents mistakes due to choosing the wrong entity to compare.

Models

NER model:

• Improved pre- and post-processing.

Decision model:

• Added ID prefix matching as an option enabled in Business Process additional parameters.

• Enabled check for blacklisted countries in a screened name (for example, ORGANIZATION "PDVSA CUBA S.A").

• Updated the name matcher to v0.2.8.

• Updated stop word lists.

• Modified the algorithm to use the location matcher only if the second name (WLE) type is Location or Default.

• Improved name matcher preprocessing for the Organization type.

Bug fixes:

• Resolved the issue with a wrong value for the country match if a country name is not set correctly and introduced additional validation to parsed countries.

• Resolved the issue with low scores for concatenated names.

  In addition to the default Tara name matching algorithm, which finds the best matching token pairs and then aggregates them to the final score, Tara now additionally generates a name pair that treats names as one concatenated token without spaces and calculate the similarity for such pairs as well.

• Resolved the issue with finding an incorrect NER entity in the hit text. Updated the multiple-hit logic is updated to avoid splitting hits if an NER entity contains all tokens present in the hit text.

• Resolved the issue with overwriting the model's location.match value and added a check for processed values.

Version 3.0.4

About Release

• Release date: May 1, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

Significant improvements to the Tara Business Process and models are designed to increase accuracy, performance, and reliability.

New features

Business Processes:

• Updated the Named Entity Recognition (NER) model to v1.6.11.

• Updated the Decision model to v4.1.5.

• Modified the getRulesFileFromS3 function to verify the existence of a file before attempting to retrieve its content.

• Updated the hit text detection logic to prioritize direct matching tokens when a watchlist entity name matches an NER entity.

• Removed BlackDuck from the Bitbucket pipeline.

Models

NER model:

• Added ABA as an identifier for the ISIN finder.

• Enhanced vessel detection by focusing on vessel names.

• Extended TEXT_LABEL by vessel-related terms, such as MT, IMO, and so on.

• Truncated long tokens to avoid performance issues.

Decision model:

• Disabled suppressing the name matching output with "NaN" due to the id_match value.

• Improved the name matcher for the Location and Organization types.

Bug fixes

Business Processes

• Fixed the SQL exception occuring when calling getAllReapplicationData() in the continuous learning flow.

• Fixed a NullPointerException in the continuous learning flow when the content format is null.

• Fixed the date-time format issue in the Continuous Learning Data Store.

• Resolved the issue with NER output parsing. If the extracted NER refers to a country name, the NER country field is populated with the related ISO2 country code.

• Fixed the index exception caused by removing newline characters from 36th positions in the tag content.

Version 3.0.3

About Release

• Release date: April 17, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8 and 10.2.9

Tara v3.0.3 focuses on an in-depth understanding of payment messages and alert details, improves disposition logic, and offers a higher automation rate.

The release provides the following additional capabilities:

• Recognition of context in composite fields and establishing linkage between elements. An example is a beneficiary field that includes an account number, a name, and an address.

• Recognition of multiple identifiers in a payment message, such as UUID, ISIN, IBAN, BIC, and other codes.

• Recognition of common language expressions in a payment message, including account types, references, and more.

• Customization of disposition logic and narratives using rules.

• Lookup, validation, and enrichment for the following external sources:

  • Google Geocoding for address validation

  • Custom sources for company, individual, or address validation

New features

• A customizable rule engine enables creating custom rules and narratives without code changes, providing final decisions and dispositions. The default rules are optimized to produce the minimum error rate.

• The Continuous Learning Data Store (CLDS) allows storing analysts' historical decisions and applying them to real-time alerts.

• The stand-alone Name Entity Recognition (NER) model enables the sending of predicted entities for enrichment and custom modifications and provides fine-tuned recognition of the ID, Individual, Organization, Location, and Vessel types.

• The unified name matcher is a reusable component you can embed into any model and use as a service.

• Recognizing a dependency between entities (for example, ID > Name > Address) increases the number of data points for accurate decision-making.

Extended entity recognition

Tara's named entity recognition now includes understanding of entity types, common language expressions, identifiers, locations, and additional information. Each recognized entity is assigned a confidence score to gradually fine-tune results and improve decision accuracy.

External source enrichment

Tara can perform external lookups and enrichment and validate information, such as location, company name, individual name, and identifiers. The enrichment functionality is optional and can be enabled as needed. Custom enrichment sources can be integrated alongside out-of-the-box options.

Rule-based disposition

The disposition logic is now rule-based. You can extend the default rule set with custom logic or modify rules to match the existing procedures. Rules also allow for narrative customization and prioritization.

Extended payment message context

Tara's context understanding capabilities are extended to account for additional payment details, such as hit types (name, embargo, ID), sender and receiver information, and the existing knowledge about entity types. When available, the data can be input into Tara to enhance disposition decisions and increase the automation rate.

For more details on new features, see the information below.

Business Processes

• Migrated Tara to Java Native Worker.

• Added the Drools rule engine to make a final decision. Created a new business object that contains all the data points, including those returned from the Decision model. A rule file is located in S3 MinIO and can be easily updated.

• Optimized the rule engine initialization.

• Added the Decision Reapplication step known as CLDS to the configuration UI.

• Enabled reading CLDS historical data from the common directory regardless of Tara's version.

• Updated CLDS historical upload statistics per run.

• Deleted the Sanctioned program and Open Corporate Connector steps from the configuraton UI.

• Enabled the application of rules on requests with errors or exceptions by allowing hits with empty tag content to go through the rule engine.

• Enabled calling to the Geocoding API only when an address is available but a country is not. The validation is displayed as true or false if Google Geocoder was used; otherwise, null is shown.

• Added hit text calculation and entity type using similarity scores and fuzzy matching. The functionality is moved from the Decision model.

• Updated joining of duplicate tag values from the content.

• Introduced the NER model v1.6.8.

• Added NER Model as a standalone bot step in the Tara Business Process. Added the nerData object for processing. NER processing is no longer part of the Decision model.

• Introduced the Decision model v4.1.13.

• Added Superset analytics into the bundle.

• Added support for multiple ID types, such as BIC, ISIN, IBAN, UUID, IFSC.

• Updated the 50F country parser and the Swift dictionary.

• Revised the stopword list for the name matcher.

• Improved Swift data parsing for Name and Address fields 50F, 50K, 50A, 59F, 59, 59A. Added multiline parsing support.

• Enabled parsing a country from IDs located in sender and receiver fields and identified entities.

• Improved country parsing and converting all countries to the ISO 2 format.

• Combined duplicate FUF tags data into a single one when parsed.

• Updated the HTML report to display a business object.

• Enabled display of the name matcher score as the reason score inside a decision object.

• Enabled an API response and analytics and a report to run in parallel, speeding up the response return to a client.

• Improved the ID matching logic.

• Added a few scenarios for parsing Additional Info per customer request.

• Added support for the ID_OTHER and ACCTYPE NER types.

• Added support for the XS country code, the ISIN special code.

• Added the isLatinChars and isHieroglyphsChars rule functions.

• Added the run_uuid and tara_version columns to the pss_request_processing_v1 Data Store.

• Added the Maximum Hits Per Message processing limit.

• Improved the logic that updates Rules Object with a parsed country and a country source.

• Added a CSV file with countries to unify country information for Rules Object.

• Implemented many changes in the rule file and utilities to decrease the error rate to less than 1%.

• Added payment information (sender, receiver, and so on) and hit type to the request JSON and processing support.

• Enabled identifying relationship between entities based on NER Name and Address, NER Name and ID, and so on.

• Added the UI configuration for the NER confidence threshold per type that allows to ignore NER predictions of low confidence.

• Updated screened data based on the Name and Address tag type and tag content. Now, the Decision model uses screened data over NER data when making a decision.

• Enabled recording of all hits with empty content to analytics.

• Added the Decision model and NER model versions to the analytics tables.

• Updated processing and reports of filtered hits.

• Added support for the VESSEL and TEXT_TAG NER types.

• Added rules for the VESSEL screened type.

• Added the logic to parse multiple IDs inside a code-map entry separated by ;.

• Added the logic to remove extraneous newline characters at 36th positions in tag content due to Swift tag length limitations.

• Set the rules service as a singleton.

• Removed the version from the DR upload directory.

• Added DR tests.

Models

• Enabled a standalone location mismatch.

• Enabled usage of the provided NER model and enrichment data for processing.

• Removed the NER dependency. Now, NER is a separate model.

• Added support for the ID type.

• Updated the logic to use data according to the provided type.

• Added the logic to compare known ID types, such as BIC, ISIN, IBAN, UUID, and so on.

• Removed the Multiple hit tokens found escalation type.

• Optimized the name natcher performance.

• Improved the name matcher logic for the Organization type.

• Added support for the NER Account and ID_OTHER types.

• Improved ID matching for the ID mismatch to return the same types only.

• Enabled the export of a processed screened name and a watchlist entity name.

• Added the pre-processing function to remove a hyphen within the ID-like values.

• Integrated unified name matcher v0.1.3.

• Enabled the output of processed names from the name matcher.

• Made the name matcher a part of the pipeline.

• Enabled the usage of the configurable name matcher threshold from the pipeline.

• Updated Python dependencies and bundled the Python zip environment to python3.11.5-pss-v2.

Bug fixes

Business Processes

• Fixed the issue with missing analytics and an HTML report for failed messages with empty tag content.

• Fixed the issue with updating the flat result Data Store.

• Fixed the issue with writing results to the flat output table.

• Fixed the issue with failing to find a tag in content when it is split into two or more parts by combining duplicate FUF tags data with count into a single one when parsed.

• Fixed performance issues when loading big historical data files.

• Fixed the issue with updating the tag content with an empty value.

• Fixed the issue with reporting duplicates to analytics tables.

• Fixed the issue with fetching a Business Process run ID in IA Cloud v10.2.7.

Models

• Fixed the issue with producing ID match and mismatch without a sanctioned ID.

• Fixed the name matcher decision when the threshold equals 1.0.

• Fixed the issue with a country incorrectly identified as a city.

• Fixed the issue with the concatenate space logic on big names for the name matcher.

• Fixed a possible Organization type mismatch when the sanctioned type is missing (equals null).

• Added various fixes for preprocessing, matchers, and decision-making.

Version 2.3.6

About Release

• Release date: April 25, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8

New features

Business Processes

• Improved hit text matching on the tag content.

• Updated a decision rule when a location and an organization are involved.

• Enabled approximate ID matching in the UI configuration.

Version 2.3.5

About Release

• Release date: March 1, 2024
• Compatible platform versions: IA Cloud 10.2.7, Work.AI 10.2.8

New features

Business Processes

• Fixed empty tag content when parsing Swift tag 50F.

• Updated the Swift dictionary.

• Updated the default Python model to v3.3.0.

• Updated country hash comparison to use only the first country in the list.

• Improved bundle packaging.

Models

• Added the unified name matcher.

• Enabled the output of the name matching score produced by the model.

• Updated a narrative for the No rules applied outcome.

• Improved libpostal output when a country was recognized as a city.

• Improved ID recognition for known types, such as BIC, IBAN, ISIN, and UUID.

Bug fixes

Models

• Fixed the outcome for the name matching threshold that was set up to 1.0.

• Resolved the "Invalid Input: nothing to repeat at position X" and "Invalid Input: missing ')'" issues.

• Resolved the issue with the organization name matching.

• Resolved the issue with the out-of-range list index.

Version 2.2.8

About Release

• Release date: September 12, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

documentation

For the documentation on version 2.2.8, refer to the Tara: PSS 2.2.6 instructions.

New features

Business Processes:

• Updated the FUF parser to combine the values of the duplicate tag content. The new parsing mechanism resolves the issue of not picking up a correct value for the cases when there are multiple sub-fields with the same name.

• Introduced Python model v3.2.11.

Version 2.2.7

About Release

• Release date: July 21, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

documentation

For the documentation on version 2.2.7, refer to the Tara: PSS 2.2.6 instructions.

New features

Business Processes:

• Updated Billing to produce a report for every transaction.

• Added Superset analytics into the bundle.

• Updated a flat input file to support CodeMap.

• Stopped overwriting filtering Data Stores during the installation.

• Introduced Python model v3.2.8.

Models:

• Improved self-service validation for predicted names with abbreviations.

• Improved "1 token rule" by allowing more resolutions for companies with legal endings.

• Improved ID matching when values are separated by spaces and punctuations.

• Stopped returning an ID mismatch when no IDs are provided on the sanctioned side.

• Stopped removing a country from a predicted name when wle_name contains it or a hit against a location.

• Optimized the country lookup.

• Improved the name mismatch procedure when a screened name matches against an ID.

Bug fixes

Business Processes:

• Fixed flow-breaking issues with saving an HTML report to S3.

Models:

• Fixed the incorrectly calculated Unable to find Hit text outcome.

Version 2.2.6

About Release

• Release date: June 29, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

New features

Business Processes:

• Added a gold decision and a gold decision comment to CSV input.

• Implemented configurable model thresholds for name match and DoB match.

• Added a billing query file to utils.

• Introduced Python model v3.2.6.

Models:

• Updated the name matcher prediction for names with apostrophes, for example, O'Connor.

• Improved predicted name matching not to consider country names.

• Improved the Multiple hit text found outcome.

• Added support for configurable DoB and name match thresholds.

Bug fixes

Models:

• Fixed a partial name mismatch result for the RESOLVE outcome.

• Fixed Type Mismatch when a sanctioned type is missing.

Version 2.2.5

About Release

• Release date: June 15, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

New features

Business Processes:

• Moved Sanction programs before Output in the configuration UI.

• Introduced Python model v3.2.5.

Models:

• Improved parsing of blacklist tokens that contain dots and spaces.

• Added the ID matching support for screened (additional) data.

• Replaced common location abbreviations with full words for better NER recognitions.

Bug fixes

Business Processes:

• Fixed the issue with the FUF parser and FircoSoft Connector Line Separator.

Models:

• Resolved the issue with the temp NER problem patch. Removed long IDs followed by a comma from the beginning of tagContent.

Version 2.2.4

About Release

• Release date: May 31, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

documentation

For the documentation on version 2.2.4, refer to the Tara: PSS 2.2.5 instructions.

New features

Business Processes:

• Added the configurable Sanction Program handling skill.

• Added the SEPA format support (similar to ISO).

• Added support for new tags in the model output.

Models:

• Added support for EO Sanction Programs.

• Introduced the ID mismatch and match decision factor.

• Enabled name mismatch categorization.

• Introduced tags in the decision output.

• Added the option to ignore middle names during name matching.

• Improved the hit text calculation and introduced the "Not able to calculate hit text" output.

• Improved additional info validation.

• Enabled display of the original screened name in the narrative.

• Improved name matching when a hit is on the number.

• Improved the usage of account holder names.

• Improved name matching when a watchlist entity name is a part of the screened name.

• Disabled Type Mismatch on Non-Entity due to NER errors.

Bug fixes

Business Processes:

• Fixed data saving (N/A) to the Analytics table.

Models:

• Fixed the Libpostal's state recognition.

• Performed a number of various bug fixes.

Version 2.2.3

About Release

• Release date: May 2, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

New features

Business Processes:

• Automated the pipeline bundle release.

• Updated Filtering Tool to allow empty and non-empty value conditions for the EXCLUDE type.

• Updated data intake for Open Corporates Calls.

• Updated a sample template file to reflect Tara's features.

Models:

• Enabled configuring the decision matrix by type.

• Added configuration for Ignoring country and currency codes in the predicted name.

• Updated the UI model configuration to the linear flow.

• Improved hit text calculation.

• Improved the Type Mismatch outcome.

• Improved blacklisted country recognition in a parsed address.

Bug fixes

Business Processes:

• Fixed GeoCoder statistics.

• Resolved various issues related to data processing.

Models:

• Fixed the issue with the _ character, where the narrative writer was stuck in a loop.

• Fixed multiple issues related to the name mismatch output.

• Resolved the issue with an invalid country name.

Version 2.2.2

About Release

• Release date: March 30, 2023
• Compatible platform version: IA Cloud 10.2.7

New features

Business Processes:

• Added the address verification skill to the configuration, flow, stats, and report.

• Improved the hierarchical structure of the HTML report location.

• Added support for multiple countries and DOB in the sanctioned data section.

• Added more scenarios to the synthetic dataset generation.

• Converted the configuration UI to the stepper flow and fixed wording.

• Updated the filtering tool to allow a custom narrative.

Models:

• Allowed the optional sanctioned entity type.

• Improved logging.

• Optimized NER.

• Removed initialization and calls of an unused BERT name matcher.

• Added support for multiple countries in the wle_country field.

• Added support for multiple DOB elements in the wle_dob field.

• Limited the number of threads for the PyTorch execution

• Allowed an optional sanctioned data type.

• Upgraded ML SDK to version 10.2.6.19.

• Improved the Type Mismatch decision of Non-Entity vs. Location.

• Updated the Type Mismatch logic to avoid E.C. classification mistakes.

• Removed and replaced [,'] characters from a narrative.

Bug fixes

Business Processes:

• Fixed the analytics report.

• Resolved the issue with saving a decision code into a Data Store.

• Resolved the issue with displaying a secret key in the Business Process output.

Models:

• Fixed exception cases. An exception only affects a hit where it occurs.

• Resolved the issue with process unique hit's tagContent.

• Resolved the issue with the geocoder validation affecting a location mismatch or match decision.

• Fixed the <location> vs. <country> Type Mismatch.

• Fixed the name mismatch on an empty name.

• Fixed the "legal ending" discovery problem.

Version 2.2.1

About Release

• Release date: February 23, 2023
• Compatible platform versions: IA Cloud 10.2.6 and 10.2.7

New features

Business Processes:

• Added the ISO20022 format parser with the limited type dictionary.

• Added Billing Report to RMQ.

• Added input and output support for connectors.

• Improved HTML Report and added an original request.

• Changed a build to produce the Tara bundle.

• Removed all Python dependencies.

• Added more scenarios to the synthetic dataset generation.

Bug fixes

Business Processes:

• Fixed the error response to return a standard response object.

• Fixed the screened data injection during content parsing for all formats.

• Resolved the issue with saving HTML reports to the analytics table.

• Applied a variety of minor bug fixes.

Models:

• Fixed the issue with hardcoded Blacklisted Configs. Now, they are read from UI configurations.

• Fixed a narrative on empty value.

• Fixed alias matching and removed the hit_not_found_in_text feature.

• Fixed Type Mismatch. NAME cannot be non_entity.

• Fixed the multiple_name_found feature.

Version 2.2.0

About Release

• Release date: January 24, 2023
• Compatible platform version: IA Cloud 10.2.6

New features

Business Processes:

• Introduced Python model v3.2.0.

• Introduced Python wheel v2.2.14.

• Updated the build pipeline.

• Updated the dashboard analytics file.

• Upgraded project dependencies, security, and the testing framework.

• Updated the flat-file input processing for FUF and SWIFT messages.

• Updated the FUF parser to get tag content.

• Packaged pre-trained models into the Tara bundle outside the PSS model.

• Moved Company Verifier to a separate bot step.

• Updated Filtering:

  • Allowed rules with only EXCLUDE conditions.

  • Enabled search for all message fields.

  • Fixed the exception using GLOBAL search.

• Updated the FUF parser to allow a newline character in a value.

• Added more scenarios to the synthetic DATA file.

Models:

• Added the screened name to the name matcher aggregator.

• Updated the matching order for aliases.

• Added country parsing from the address field.

• Replaced the predicted name with an account holder's name when applicable.

• Added the hit_not_found_in_text feature.

• Added additionalData_name versus predicted_name validation.

• Reworked decision narratives to include the NO_DECISION explanations.

• Changed the project build by moving heavy pre-trained modules to the Python installation.

• Applied trusting user's data when calculating the predicted name.

• Updated the name matcher:

  • Added checking the screened name before the account holder's name.

  • Fixed the multi-alias matching logic.

• Enabled treating the \n character as a separator for the NER text processing.

Other:

• Re-structured Tara. Now, the initial installation is 5 GB, with each subsequent update around ~200 MB.

Bug fixes

Business Processes:

• Fixed the FUF dictionary file.

• Fixed the narrative processing and display.

Models:

• Fixed predicted type_mismatch for one token named.

• Fixed the additionalData type handling.

• Fixed the primary name aggregation issue.

• Amended type_mismatch for predicted names.

• Fixed various bug fixes related to making a decision.

Version 2.1.5

About Release

• Release date: November 23, 2022
• Compatible platform versions: IA Cloud 10.2.5 and 10.2.6

New features

Business Processes:

• Added the Open Corporates usage analytics.

• Enabled support for SWIFT messages as input.

Models:

• Added two type_mismatch outcomes: hit on location and hit on other word.

• Allowed resolving hits with one wle_name token to match an old model.

• Improved the N.M. token overlap check.

• Updated the partial date comparison.

• Enabled considering the type input when making a Type Mismatch decision.

• Synced up black and stop lists with the WISE model.

• Added currency and country filters to N.M. preprocessing.

• Migrated missing test scenarios.

Bug fixes

Business Processes:

• Fixed the message decision after an Open Corporates call.

Models:

• Resolved the N.M. issue with the punctuation in wle_name.

• Fixed the issue with catching ValueError during input validation and returning NO_DECISION.

Version 2.1.4

About Release

• Release date: November 15, 2022
• Compatible platform versions: IA Cloud 10.2.5 and 10.2.6

New features:

• Introduced Python wheel v2.2.10.

• Upgraded the ODF 2 and WorkFusion dependencies.

• Introduced Bot Task integration tests.

• Added the flat file explanation document.

• Updated the configuration for Open Corporates and template download.

• Updated integration tests and synthetic data generation.

• Allowed supplying the Open Corporates license and creation of Secrets Vault programmatically.

Version 2.1.3

About Release

Compatible platform versions: IA Cloud 10.2.5 and 10.2.6

New features:

• Introduced Python wheel v2.2.9 and 3.0.2.

• Enabled the FUF format processing.

• Added missing fields to the filtering tool.

• Updated type handling from the flat file.

• Applied fixes for the payload configuration and form, Tara packaging, and data filtering.

Version 2.1.1

About Release

Compatible platform version: IA Cloud 10.2.5

New features:

• Enabled reading configuration from a JSON file during the Prepare Use Case phase.

• Enabled data processing from WorkFusion API.

• Removed Data Store monitoring. Now, you can provide data via File Upload or send through API.

• Integrated a new standard model.

• Updated filtering rules to allow the EXCLUDE records.

• Updated testing scenarios, the DW test framework, and performance tests.

• Updated the DW model result processing.

• Migrated the SWIFT and FUF datasets to the standard Tara input format.

• Added the DW full bundle packaging for Solution Catalog.

• Updated analytics tables and HTML report generation.

Version 2.0.0

About Release

• Release date: May 20, 2022
• Compatible platform version: IA Cloud 10.2.4

New features:

• Migrated to Liquibase.

• Migrated to ODF 2. Now, Business Processes are compatible with IA Cloud v10.2.4.

• Added support for file upload from the Control Tower data input.

• Added column indexing for request and analytics Data Stores.

• Enabled using the BitBucket pipeline for building, deploying, and testing.

• Updated the HTML report generation.

• Updated integration tests.

• Initiated preparation to use the Prepare Use Case configurations.