Overview

A movie recommendation service built like a real product, not a notebook: separated ingestion, feature engineering, training, evaluation, and serving layers over MovieLens interaction data and item metadata. The model is a hybrid, content-based similarity, item co-occurrence signals, and popularity bias for cold-start stability, served through FastAPI endpoints for personalized recommendations, similar-item retrieval, health checks, and hot model reloading.

Highlights

• Hybrid recommender: content similarity + co-occurrence + popularity prior
• FastAPI serving layer with API-key protected endpoints
• Docker Compose setup, optional Redis caching, structured logging, request tracing
• CI-ready testing workflows and modular train/serve pipelines

Stack: Python · FastAPI · scikit-learn · Airflow · Redis · Docker

Links

• Code: github.com/saaim12