{
  "id": 9646968,
  "title": "JEVelric : An Agentic Fraud Investigation System on TigerGraph",
  "url": "https://urgent.news/2026/09/24/jevelric-an-agentic-fraud-investigation-system-on-tigergraph",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-24T22:47:45.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/tanmay_sayare_c95b00a974e/jevelric-an-agentic-fraud-investigation-system-on-tigergraph-3kbm"
  },
  "original_language": "en",
  "account": "JEVelric is an AI agent designed to investigate card fraud in a manner similar to human fraud analysts. It is triggered by risk scores, customer complaints, or analyst requests, and retrieves relevant evidence from a knowledge graph to assess confidence levels. The agent may request additional evidence if it is not confident enough to make a decision, but ultimately relies on a deterministic policy engine to recommend actions. These actions are linked to specific policy rules, ensuring regulatory compliance and explainability. JEVelric was developed as part of the TigerGraph's HHGOA Agentic Fraud Investigation hackathon, utilizing the IEEE-CIS/Vesta dataset as a benchmark. The system employs a TigerGraph database to store relationships between cardholders, cards, transactions, device profiles, email domains, billing regions, and closed cases. The agent interfaces with the graph database through the TigerGraph MCP, which handles schema creation, evidence retrieval, and case writing. JEVelric's key features include uncertainty-gated evidence gathering, policy-governed actions with approval routes, and a live investigation endpoint for processing new cases. However, the main challenges faced during development included ensuring the trustworthiness of the graph integration and dealing with free-tier LLM rate limits. Future improvements could involve implementing a full TigerGraph vector-backed retrieval system and adding a human-in-the-loop approval UI for certain actions.",
  "summary": "What we built JEVelric is an AI agent that investigates card fraud the way a human fraud analyst would: it takes a trigger — a risk score, a customer complaint, or an analyst request — pulls the actual evidence from a knowledge graph, weighs how confident it can be, asks for more evidence when it can't be confident yet, and only then recommends an action — citing exactly which policy rule…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Dev.to",
        "title": "Building an Agentic Fraud Investigation System with TigerGraph, LangGraph & MCP",
        "url": "https://urgent.news/2026/09/24/building-an-agentic-fraud-investigation-system-with-tigergraph",
        "published": "2026-09-24T14:56:24.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}