Reginx is short for recommendation engine X. I plan to build most part of modern recommendation engine from scratch, initial plan including:
- Popular machine learning models like CF, FM, XGBoost, TwoTower, W&D, DeepFM, MaskNet, SASRec, Bert4Rec, Transformer, etc.
- Online inference service written by Golang, including candidate generator, ranking and re-ranking layers
- Feature engineering and preprocessing, including both online and offline part
- Diversity approaches, like MMR, DPP
- Deduplication approaches, like LSH or BloomFilter
- Training data pipeline
- Model registry, monitoring and versioning