Back to projects
Machine Learning

Traffic Anomaly Detection System

End-to-end MLOps pipeline for anomaly detection on traffic sensor data, using MLflow and DVC for experiment tracking and reproducibility, model serving containerized with FastAPI and Docker, and Prometheus/Grafana monitoring for real-time metrics and observability.

  • MLOps
  • MLflow
  • DVC
  • FastAPI
  • Docker
  • Prometheus
  • Grafana

The infrastructure around the model

The focus here is what surrounds the anomaly-detection model: reproducible training, containerized serving, and production observability, wired together end to end.

Sensor DataTraining (MLflow + DVC)Serving (FastAPI + Docker)Prometheus / Grafana

Tracking and reproducibility

  • MLflow tracks experiments, parameters, and metrics across training runs.
  • DVC versions the traffic sensor datasets and model artifacts so a run can be reproduced from the exact data it was trained on.

Serving and observability

  • The trained model is containerized and served through FastAPI and Docker.
  • Prometheus scrapes serving metrics and Grafana dashboards them for real-time observability and performance tracking in production.