Movie Recommender
Hybrid ML engine combining collaborative filtering and content-based models, benchmarked across five algorithms.
March 2026 – May 2026
Overview
A deep dive into how recommendation systems actually behave in production-like conditions. Rather than picking one algorithm and tuning it, I built a rigorous evaluation framework to fairly compare five different approaches on the same dataset — exposing the real tradeoffs between accuracy, coverage, and cold-start performance.
My Role
Designed and implemented all five models, built the shared evaluation pipeline from scratch, and wrote the final analysis comparing their tradeoffs. The hybrid model was the main deliverable — the others exist specifically to give it something meaningful to beat.
Approach
Used SVD-based collaborative filtering as the foundation, then layered in content-based signals to handle cold-start cases. Tuned hyperparameters via grid search and validated generalization with k-fold cross-validation against a simple popularity baseline. RMSE landed at 0.94 on the held-out set.
What I'd Do Differently
TODO: What would you do differently on this project?