Application GB2602661.7

REPLACEMENT DESCRIPTION — APPLICATION GB2602661.7

TITLE OF INVENTION

AI-ORCHESTRATED MULTI-SOURCE BACKGROUND SCREENING AND COMPLIANCE MANAGEMENT SYSTEM WITH REAL-TIME SANCTIONS DETECTION, ADVERSE MEDIA ANALYSIS, AND RISK-SCORED SUBJECT INTELLIGENCE REPORTING

NOTE: This replacement description is submitted in response to the Intellectual Property Office examination letter dated 03 June 2026 under Section 15A of the Patents Act 1977. This description supersedes the description previously filed with application GB2602661.7.

TECHNICAL FIELD

The present invention relates to artificial intelligence systems applied to due diligence, compliance, and background screening. More specifically, the invention concerns: AI-orchestrated multi-source screening systems that simultaneously query criminal records, sanctions databases, PEP lists, credit registries, and adverse media sources; machine learning models for automated subject risk scoring and entity resolution across multiple jurisdictions; natural language processing systems for adverse media classification and risk signal extraction; and multi-tenancy compliance management platforms. The invention falls within the technical fields of: information retrieval and indexing (IPC G06F 16/903); machine learning (IPC G06N 20/00); business process management and compliance (IPC G06Q 10/06); and public safety and law enforcement systems (IPC G06Q 50/26).

BACKGROUND OF THE INVENTION

The background screening and compliance sector is characterised by fragmented tooling, manual processing bottlenecks, and an absence of AI-driven risk aggregation.

1.1 Background Screening Systems
Existing background screening platforms (Checkr, First Advantage, HireRight, Sterling) operate as sequential, single-source query systems. Each data source is queried independently, results are delivered as raw text without automated risk assessment, and no system provides a unified risk score with per-factor SHAP explainability. No existing commercial platform simultaneously orchestrates 50+ global criminal record databases, sanctions lists, PEP registries, adverse media crawlers, and corporate registry checks via a single API call with normalised output.

1.2 Sanctions Screening
Existing sanctions screening tools (World-Check, Refinitiv, Dow Jones Risk & Compliance) screen against individual lists but do not perform cross-list deduplication, do not generate phonetic/transliteration variants for non-Latin script names, and do not provide explainable match confidence scores indicating which name variant matched which list entry.

1.3 Adverse Media Analysis
Existing adverse media tools rely on keyword searches returning unprocessed news articles. No system applies NLP classification to categorise adverse media findings by risk category (fraud, corruption, criminal, regulatory, litigation), nor to filter irrelevant results through entity disambiguation.

1.4 Risk Score Aggregation
Background screening results are currently delivered as separate data items without a unified risk score. Compliance officers must manually weigh findings from multiple sources. No commercially deployed system applies an explainable machine learning model to produce a single risk score with per-factor contribution explanations suitable for audit trail purposes.

SUMMARY OF THE INVENTION

According to a first aspect of the present invention, there is provided a computer-implemented AI-orchestrated background screening and compliance management platform comprising five integrated components: (a) an AI orchestration engine and subject risk scoring system; (b) a multi-source data integration architecture covering 50+ global data sources; (c) a real-time sanctions and PEP screening engine with variant generation; (d) an adverse media NLP classification and risk signal extraction system; and (e) a SHAP-explainable risk aggregation model, further comprising a multi-tenancy client portal and case management system as set out in the claims below.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention. All drawings are in black and white.

Figure 1 is a block diagram showing the overall AI-orchestrated platform architecture.
Figure 2 is a flowchart illustrating the AI orchestration engine and subject risk scoring pipeline (Component A).
Figure 3 is a modular adapter diagram showing the multi-source data integration architecture (Component B).
Figure 4 is a layered pipeline diagram showing the real-time sanctions and PEP screening engine (Component C).
Figure 5 is a parallel pipeline diagram showing the adverse media NLP classification system (Component D).
Figure 6 is a SHAP waterfall chart showing the risk score aggregation and feature contribution output (Component E).
Figure 7 is a flowchart showing the client portal workflow from case initiation to report delivery.
Figure 8 is a jurisdiction connector map showing multi-jurisdiction criminal record coverage.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

1. SYSTEM ARCHITECTURE OVERVIEW (with reference to Figure 1)

The platform is implemented as a cloud-deployed, API-first, multi-tenancy SaaS system. As shown in Figure 1, the primary components are: a Client Interface Layer (10, 20) comprising separate B2B enterprise and B2C self-service portals; an AI Orchestration Engine (30) comprising five integrated components (31–35); a Data Integration Layer (40) comprising modular adapters to 50+ external data sources; and a Storage Layer (50) comprising relational, vector, and audit log databases.

2. AI ORCHESTRATION ENGINE — COMPONENT A (with reference to Figure 2)

The AI orchestration engine (31) is implemented as shown in Figure 2. Upon receipt of a subject intake request (100) containing name, date of birth, identifier, and country, the orchestrator (110) dispatches parallel queries to all configured data source adapters. Results are normalised (120) into a unified schema, entity resolution (130) confirms match confidence, and the risk scoring module (140) produces a score from 0–100 with per-factor contribution. The report generator (150) produces the structured subject intelligence report.

3. MULTI-SOURCE DATA INTEGRATION — COMPONENT B (with reference to Figure 3)

The data integration architecture (32) as illustrated in Figure 3 provides a unified abstraction interface (200) over six adapter categories (210–215): Sanctions (OFAC/UN/EU/HMT), Criminal Records (50 jurisdictions), Credit/Financial registries, Corporate registries, Adverse Media crawlers, and PEP databases. Each adapter (220) normalises source-specific formats to the platform's unified schema.

4. SANCTIONS AND PEP SCREENING ENGINE — COMPONENT C (with reference to Figure 4)

The sanctions screening engine (33) as illustrated in Figure 4 takes subject name and identifiers as input. A variant generator (300) produces phonetic, transliteration, and alias variants. Parallel queries (310) are dispatched to OFAC SDN, UN Consolidated, EU Consolidated, HM Treasury, and INTERPOL lists simultaneously. A fuzzy match scorer (320) computes match confidence. The decision engine (330) applies configurable thresholds to produce CLEAR, POSSIBLE MATCH, or CONFIRMED MATCH outputs. Unresolved cases are flagged (340) for analyst review.

5. ADVERSE MEDIA NLP SYSTEM — COMPONENT D (with reference to Figure 5)

The adverse media system (34) as illustrated in Figure 5 aggregates results from web crawlers and news APIs (400). An NLP classifier (410) categorises findings by risk type: fraud, corruption, criminal, regulatory, and litigation. A sentiment module (420) scores negativity. A relevance filter (430) applies entity disambiguation to remove false positives. The risk tier output (440) classifies adverse media as None, Low, Medium, High, or Critical.

6. RISK SCORE AGGREGATION — COMPONENT E (with reference to Figure 6)

The risk aggregation model (35) employs a gradient boosted ensemble with SHAP explainability. As illustrated in Figure 6, per-factor SHAP contributions (510, 520) from a base score (500) produce the final subject risk score (530). Positive contributors include sanctions hits, criminal records, adverse media tier, and PEP status. Negative adjustments include verified residency history and employment verification. The SHAP output (540) provides an audit-ready explanation for each score.

7. CLIENT PORTAL WORKFLOW (with reference to Figure 7)

As illustrated in Figure 7, the client workflow (600) proceeds from case initiation through AI orchestration (610) to analyst review (620) and report delivery (630). The platform supports both automated instant checks and enhanced due diligence cases requiring human oversight.

8. JURISDICTION CONNECTOR MAP (with reference to Figure 8)

As illustrated in Figure 8, a Unified Query Interface (700) routes requests to regional connector categories (710–715) covering UK/Ireland, Europe, North America, Middle East/GCC, South/Southeast Asia, and Sub-Saharan Africa. An authentication handler layer (720) manages API Key, OAuth 2.0, Certificate, and Session Token authentication per data source.

CLAIMS

[Claims remain as filed — see original claims document filed with application GB2602661.7. Claims are not amended by this replacement description.]

ABSTRACT

A computer-implemented AI-orchestrated platform and method for multi-source background screening and compliance management. The platform integrates: (i) an AI orchestration engine that simultaneously queries 50+ criminal record databases, sanctions lists, PEP registries, adverse media sources, and corporate registries via a single API call; (ii) a modular adapter architecture enabling jurisdiction-specific data source integration without core system modification; (iii) a real-time sanctions and PEP screening engine with phonetic and transliteration variant generation and fuzzy match scoring; (iv) an NLP-based adverse media classification system categorising risk signals by type and severity; and (v) a SHAP-explainable risk aggregation model producing unified subject risk scores with per-factor contribution breakdowns suitable for compliance audit purposes. The system serves B2B clients including background screening firms, banks, law firms, and HR departments, and provides multi-tenancy case management with full audit trail, consent management, and configurable risk thresholds.