AI / ML/2026/Live
IntelliFraud Copilot
A copilot for fraud analysts. Several LLM agents work a case together, pulling from policy documents with hybrid search, watching transactions as they land, and running the AML/KYC checks. Whatever it concludes comes with a plain explanation an analyst can actually sign off on.
Context
Fraud analysts drown in alerts and thick policy manuals. IntelliFraud sits alongside them to move an investigation along faster, and just as importantly it shows its reasoning, so a person can sign off with confidence rather than trust a black box.
Approach
- 01Built a multi-agent LLM architecture where specialized agents handle distinct investigation tasks.
- 02Grounded answers in AML/KYC policy with hybrid retrieval — vector plus keyword search over a Qdrant index.
- 03Monitored transactions in real time and produced explainable, analyst-facing summaries rather than opaque scores.
- 04Backed it with PostgreSQL and a React investigation interface.
Notable
- Live and reachable at fraud-detector.africodelab.net.
- Explainable summaries keep a human analyst in the loop.
- Compliance-aware retrieval over real policy documents.
Next project
RideSync+