Albedo is an open-source recommender system aimed at helping developers discover GitHub repositories by learning from activity signals. It treats repositories and developers as a graph of interactions and applies large-scale matrix factorization to model affinities, with Apache Spark providing the distributed data processing. The project focuses on implicit feedback—stars, watches, and other engagement metrics—so it can build useful recommendations without explicit ratings. A reproducible setup and Makefile-driven workflow streamline tasks like spinning up services, loading datasets, training models, and generating candidate lists. Because it’s built around Spark’s scalable primitives, Albedo can experiment on substantial snapshots of GitHub metadata rather than toy corpora. The repo is also educational: it demonstrates a practical end-to-end pipeline from ingestion and feature preparation to training and ranking.
Features
- Spark-based pipelines for scalable training and inference
- Implicit feedback modeling for stars, watches, and similar signals
- Reproducible Makefile commands for setup, training, and evaluation
- Configurable ALS or similar factorization models for recommendations
- Batch generation of ranked repository suggestions per user
- Modular project structure suited to experimentation and extension
Project Samples
