Enterprise Banking Fraud Monitoring Platform

Overview

A simulation of how a bank or FinTech would monitor transactions for fraud. Every transaction is scored twice: a Spring Boot rule engine runs seven pluggable rules, including cross-customer fraud-ring detection over shared device IDs and IP addresses, then calls a FastAPI service for an Isolation Forest anomaly score, and the two are combined into a single result. If the ML service is unreachable a circuit breaker trips and scoring continues on rules alone, so fraud scoring never blocks a transaction. Transactions arrive either over REST or through Kafka-compatible Redpanda streaming, with a dead-letter topic for anything that fails. Alerts feed an analyst workflow — assignment, bulk operations, escalation, SLA policies with breach tracking, and investigation cases with their own activity timeline — while customer records carry PII encryption at rest, an aggregated risk view and a lock-and-unlock workflow that requires admin approval. Every state change is written to an immutable audit trail.

What it does

• Hybrid scoring: a seven-rule engine combined with an Isolation Forest anomaly score
• Fraud-ring detection linking customers by shared device ID or IP address
• Circuit breaker that falls back to rule-only scoring when the ML service is unreachable
• Streaming ingestion over Kafka-compatible Redpanda with a dead-letter topic
• Analyst workflow with alert assignment, bulk operations, escalation and investigation cases
• SLA policies with breach and near-breach tracking
• Customer risk view, PII encryption at rest and an approval-gated account lock workflow
• Immutable audit trail behind every state change