About this project
Madoff is a production-ready MCP (Model Context Protocol) server that automates end-to-end insurance claim fraud detection using a multi-layer AI pipeline. When a claim is submitted, Madoff runs a 5-stage investigation pipeline autonomously: 1. Document Retrieval — pulls the claim and its supporting images from Cloudinary 2. Vision AI Analysis — uses Groq's Qwen 3.6 multimodal model to visually inspect document images and cross-check them against the claim text (detecting forged receipts, mismatched license plates, stock photos used as evidence) 3. OCR Extraction — extracts raw text from documents for structured audit trails 4. Rule Engine Scoring — applies deterministic risk rules (duplicate claims, geolocation anomalies, transaction velocity) to generate a 0–1 risk score 5. Final Decision — combines AI confidence with rule scores to auto-approve, reject, or escalate to human review Built with NitroStack MCP framework, MongoDB Atlas for claim storage, and Groq Cloud for AI inference. Deployed on NitroCloud with both HTTP and STDIO transport — making it compatible with any MCP client including Claude Desktop, Cursor, and Copilot. The system exposes 11 structured tools including analyze_claim, freeze_account, execute_kyc, check_duplicate_claims, and generate_investigation_report — all callable by any AI agent or human investigator through a unified MCP interface. Real-world impact: catches fraud patterns that rule-based systems miss, including visually forged documents, cross-border claim inconsistencies, and coordinated fraud rings — all without a single human reviewing the initial claim.
Enterprise AI & Workplace Automation track
Develop AI agents and automation tools that improve productivity, streamline workflows, and enhance business operations.