API Guardian — ML-Powered Adaptive API Protection Platform

API Guardian — ML-Powered Adaptive API Protection Platform (Graduation Project, team of 2)

• Built a real-time protection layer where Envoy Proxy checks every request against a custom Python/FastAPI decision

service over gRPC before allowing it through.

• Combined an offline Gradient Boosting classifier with a continuously-learning online model (HalfSpaceTrees / River) so

the system adapts to new abuse patterns without manual retraining.

• Built full observability into the system — Prometheus metrics and Grafana dashboards with automated alerts on

abnormal block rates and risk scores.

• Diagnosed and fixed real production-style failures: bot-driven model poisoning, false-positive blocking, lost model state

on restart, and proxy misconfiguration.

• Orchestrated the full stack (proxy, ML service, Redis, monitoring) with Docker Compose on a Linux VM.