{
  "id": 9473824,
  "title": "FraudLens: We Built a Fraud Investigator on TigerGraph, Not a Fraud Classifier",
  "url": "https://urgent.news/2026/09/24/fraudlens-we-built-a-fraud-investigator-on-tigergraph-not-a-fraud",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-09-24T03:05:21.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/yashforsure/fraudlens-we-built-a-fraud-investigator-on-tigergraph-not-a-fraud-classifier-3knc"
  },
  "original_language": "en",
  "account": "FraudLens developers avoided creating a fraud classifier by focusing on relationships instead of scoring individual transactions. The system is built on a graph database, TigerGraph, allowing it to investigate alerts and gather evidence from the relationships between customers, cards, transactions, devices and regions. The investigators ask questions such as whether the transaction fits the customer's spending history, if the device has been used before, and whether there's a more innocent explanation for the alert.\n\nFraudLens compares the alert to historical closed-case investigations in TigerGraph, and estimates the probability and uncertainty of fraud, before making an initial decision. If further evidence could change the outcome, the system requests more information from the customer. The final decision is made by a deterministic policy engine that determines whether an action, such as blocking a card, requires approval from an analyst. The entire process is logged back into TigerGraph for confirmation.",
  "summary": "FraudLens: building a fraud investigator on TigerGraph, not a fraud classifier By Team TrustMeBro (Yash, Manan, Priyank), built at Goa Hackerhouse 2026 with @TigerGraphDB and @247pmstudio Why we didn't build a classifier Before Goa Hackerhouse 2026, none of us had used a graph database seriously. The challenge changed that. We were given a bank's card transactions, a history of closed fraud…",
  "key_points": [
    "FraudLens built on TigerGraph graph database for investigating alerts",
    "System asks questions about transaction fit, device usage, and explanations",
    "Final decision made by policy engine, requiring analyst approval if needed"
  ],
  "editors_take": "This approach shifts the focus of fraud investigation from automated scoring to a more nuanced, evidence-based inquiry, potentially leading to more accurate and fair outcomes.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "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."
}