Stack
React, PyTorch, scikit-learn, FastAPI, Python, Firebase, Tailwind CSS
Background
Chess Trainer came to life after watching grandmasters compete in tournaments and becoming obsessed with the way they think. I wanted to improve my own strategic play and track how fast I was actually getting better over time. The most natural way to measure that felt like training a model on my own Chess.com games, a CNN that could mimic my playstyle and give me something real to play against. Probably could have just used Stockfish at a lower difficulty, but where's the fun in that?
Overview
Chess Trainer is a full-stack chess learning platform with two core modes: a tactical puzzle trainer with 1,000+ puzzles and progress tracking via Firestore, and an AI opponent trained on 2,600+ personal Chess.com games. The AI uses a convolutional neural network with board positions encoded as 8x8x12 bitboard tensors fed through 3 conv layers and 2 fully connected layers to replicate a specific player's style, making every game feel like playing against a real human rather than a generic engine. Users sign in with Google OAuth, keeping progress and stats tied to their account across sessions.