Real-Time Fraud & Statutory AML.
A GCP-native modular monolith delivering sub-150ms transaction scoring, in-process ML inference, and automated regulatory reporting for modern financial institutions.
Right-Sized Architecture (PRD v3.1)
Instead of paying fixed compute overhead for idle prediction endpoints and complex message brokers, Vigil runs in-process ML model scoring inside the Core API and stores feature vectors in Cloud Firestore on demand.
- In-Process ML InferenceSUB-150MS SLA
Trained XGBoost and LightGBM models served directly in-process within the Core API memory space. Eliminates Vertex AI prediction endpoint network hops and fixed monthly costs.
- Statutory Compliance LayerCBN / NFIU 2026
Automatically converts detection events into statutory STR (Suspicious Transaction Report) and CTR XML payloads compliant with the CBN Automated-AML circular and NFIU directives.
- Modular Licensing EngineFIRESTORE BACKED
Enterprise licensing system allowing dynamic per-module activation (Fraud vs. AML vs. Sanctions/PEP) without code changes or service redeployments.
- Scale-to-Zero ArchitecturePAY-AS-YOU-GO
POC and Free tiers scale to zero instances when idle. Paid SLA customers operate with warm instances, ensuring cost control without compromising sub-150ms bounds.
CBN & NFIU Regulatory Automation
NFIU STR Filing
Generates automated Suspicious Transaction Report XML payloads conforming to NFIU submission schema with full audit traceability.
CTR Automation
Aggregates cross-channel cash transactions exceeding regulatory thresholds (Individual & Corporate) for statutory filing.
In-Line Screening
Screens inbound and outbound parties against OFAC, UN, and local watchlists during the transaction authorization window.
SHAP Explanations
Outputs feature contribution weights for every flagged event, giving fraud review teams clear reasoning for every block.