prashanna080a-tech/Agentic-fraud-detection — reverse-engineered prompt

Build me a working FraudShield demo for real time transaction fraud detection. I want a simple web dashboard and API where I can submit sample eSewa style transactions and immediately see whether each one should pass, ask for OTP, or be blocked.

Use the idea of five fraud checks, transaction speed and spending velocity, impossible travel based on location, user behavior compared to normal habits, suspicious account connections in a graph, and recent transaction sequence patterns. Combine their scores in an orchestrator so remittance, POS, and QR payments can be weighted differently. Show the final score, the agent scores, the decision, and the reason in plain language.

Include sample users like Sita and Ram, synthetic transaction data, and demo cases for normal payments, borderline OTP cases, and obvious fraud. If a transaction needs OTP, send both SMS and email if credentials are available, otherwise mock it cleanly. Use Flask for the app, and connect Redis, Kafka, and Neo4j if they’re running, but make the demo still usable locally with clear fallbacks. Look up current docs online if needed.