FraudLens AI: An Autonomous Graph Agent That Investigates Financial Crime
Fraud rarely looks like fraud in a single row of data. It shows up in the connections: a shared IP address, a reused device, a chain of accounts that all trace back to one bad actor. Investigators still find those connections by hand, cross-referencing CSVs and relational tables. One synthetic identity cluster can take hours to untangle, and by the time the ring is mapped, the money has moved.…
Financial crime often goes undetected because it appears as a complex web of connections between accounts, devices, and IP addresses. Traditional methods involve manually cross-referencing data, which is time-consuming and only identifies the connections after the money has already been moved. To address this issue, FraudLens AI was created as an autonomous system that can investigate suspicious transactions on its own.
FraudLens AI is an agentic triage system that autonomously investigates flagged transactions, maps the multi-hop blast radius of a threat inside TigerGraph, decides whether to autonomously freeze the account or require human approval, generates a compliance-ready Suspicious Activity Report (SAR) as a downloadable PDF, and demonstrates the entire process via a React dashboard with live updates.
The system consists of four layers:
1. A React + Tailwind dashboard for the investigator's workspace.
2. A Django backend that streams the agent's reasoning live over Server-Sent Events.
3. A LangGraph agent that runs the investigation loop.
4. TigerGraph, which serves as both the knowledge engine and the long-term case memory.
The investigation process begins with the ML Fraud Model flagging a high-risk transaction. The agent then calls its tools - Blast Radius to pull connected entities from TigerGraph and Policy Check to verify the case against internal fraud policy rules. It reasons over the data, citing actual entities and relationships within the graph database, and chooses the next best action.
If the agent can act independently, it freezes the account (L1 Execution). If human judgment is required, it escalates the case (L2 Approval). The system generates a PDF SAR containing the findings and supporting evidence.
The key learning from this project is that while LLMs can reason fluently, they lack grounding without real graph paths from TigerGraph. By integrating TigerGraph as the knowledge engine and long-term case memory, FraudLens AI ensures that its claims are based on factual data and not fabricated connections. The system also avoids the slower, more error-prone process of relational database joins by using a graph database to traverse relationships efficiently.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
