Recommender Systems · Conversational AI · Full-Stack
BookDB blends collaborative filtering, semantic search, and a conversational AI. Tell the chatbot what you're in the mood for and get personalised picks ranked by several ML models — fused with weighted scoring and RRF reranking — plus the reasons why, drawn from real reviews.
↑ A scripted illustration of the chat recommender — results are from a small sample catalog.
01 — How it works
Unlike engines that recycle the same bestsellers, BookDB learns from your reading history and taste profile. Several strategies run in parallel and their outputs are fused for the most relevant result.
BPR, SAR, and NCF models trained on reading interactions.
Vector similarity over book embeddings for "books like this".
Weighted scoring and reciprocal-rank fusion combine every signal.
An LLM rewrites your query for retrieval and routes tools (search, recommend, review RAG).
Tells you why it recommends a book, pulling from real user reviews.
A Go-based Model Context Protocol server exposes tools to Claude Desktop, Cursor, etc.
02 — Scale
03 — Stack