Movie Recommendation Project
Hybrid recommender system and competition project
- Status
- Competition entry
- Engagement
- Programme final project (optional)
- Role
- Sole ML engineer
The problem
Viewers need personalised movie recommendations, and public-leaderboard scores do not translate directly to real-world KPIs.
What was built
A hybrid content-based and collaborative-filtering recommender built on about 10 million ratings, using SVD matrix factorisation and LightGBM. Public-leaderboard RMSE improved from 1.21 to 0.82 across eight submissions.
Architecture and engineering decisions
Python data preparation and feature engineering over roughly 10 million ratings, a 50-factor SVD representation, LightGBM prediction, and offline evaluation against the competition’s public leaderboard.
Audience
Portfolio and competition demonstration; not a commercial product.
Technical highlights
- 50-factor SVD matrix factorisation
- LightGBM prediction layer
- About 10 million ratings
- Public-leaderboard RMSE 1.21 → 0.82 across eight submissions
My responsibilities
- Data preparation
- Feature engineering
- Model training
- Evaluation and submissions
Why it matters for a client
Shows measured model iteration and honest evaluation boundaries for a data/ML engagement.
Constraints and trade-offs
Public-leaderboard metric only; do not extrapolate to customer KPIs or production adoption.
Privacy and accessibility
Uses a public movie-rating dataset; no personal user data.
Verified evidence
Public source at github.com/rayklanderman/mrp-2026. Public-leaderboard RMSE 1.21 → 0.82 across eight submissions, as reported on the owner portfolio.
Limitations
Competition and prototype scope; not a production recommender; leaderboard context only.