AI Chess Mentor Pro
Turns raw chess-engine output into coaching a human can actually learn from.
Role
Solo build — retrieval pipeline, engine integration, and interface.
Context
Engines like Stockfish tell you the best move, not why it's the best move. For someone trying to improve, a bare evaluation number isn't useful — it's the idea behind the move that teaches you something.
I play chess seriously myself, so this started as a tool I wanted to use, not a market I researched first.
Approach
Stockfish evaluates the position; a retrieval pipeline over opening theory and pattern data (pgvector) pulls the relevant context for that exact position.
The model takes the engine's output and the retrieved theory and explains the idea in plain language — the plan, the threat, the trade-off — instead of just naming the move.
The frontend (React, GSAP) is built around a board, not a chat window, so the explanation stays attached to the position it's about.
Outcome
I use it for my own training, and it's now a working demo of retrieval-augmented coaching rather than retrieval-augmented search.
Stack
Python, FastAPI, Stockfish, pgvector, OpenAI, React, GSAP