Hybrid Recommendation System on Amazon Reviews 2023
Architected the vector-search and unsupervised-learning engine behind a hybrid recommender over 4.5M+ reviews and 137K products, with TF-IDF, Item2Vec and SBERT embeddings for semantic search and ranking. Built FAISS and ChromaDB HNSW indexes, combined SVD with content signals for re-ranking, and grouped 36K+ users into 10 personas (silhouette 0.832).