ChessMind is born from a frustration every chess player knows: engines like Stockfish give you the perfect move, but never explain WHY. And when you try to understand the refutation, you realize you need to calculate 20 moves ahead and sacrifice a piece — useless for a 1500-rated player.
ChessMind solves this by combining three pillars:
1. Computer vision (YOLO) to detect a physical board via a webcam (offline, no cheating).
2. Lichess API to analyze your FINISHED games (never real‑time, to keep it ethical).
3. A smart database of middle‑game positions, extracted from YOUR own game history, not from random masters.
Here’s how it works:
· It downloads your last 100 games for a given opening (say, Spanish or Sicilian).
· It uses clustering (unsupervised learning) to group similar middle‑game positions that you face most often.
· For each cluster, it runs Stockfish to find the best moves and calculates the 'error depth' — how many moves the engine needs to punish a sub‑optimal move (so you know if it’s a human‑level mistake or a computer‑level blunder).
· It then presents you with a clear PLAN in natural language (e.g., 'Pressure the e‑file with rooks and push the d‑pawn'), generated either by a small language model or by Wingman itself.
· Finally, it builds a spaced‑repetition schedule (like Anki) to help you internalize these plans over time, exactly when you need them.
Why 5‑6 openings? Because I want to cover the most common ones for club players. For each opening, I’ll prepare ~20 middle‑game positions per rating level (1500, 1800, 2200, 2500+). At lower levels, positions are more diverse; at higher levels, they converge, making the system even more effective.
Technology stack:
· YOLO for board detection (AI vision).
· Lichess API for game history.
· Stockfish for move validation and error‑depth analysis.
· Clustering (K‑means) to group similar positions (unsupervised learning).
· A lightweight LLM or Wingman pipelines to generate human‑readable plans.
· All runs locally on Axelera’s hardware, with zero cloud dependency.
4‑week plan:
· Week 1: Set up webcam + YOLO, integrate Lichess API, download game history.
· Week 2: Extract middle‑game positions, apply clustering, create a database of 20 positions per opening (5‑6 openings) with Stockfish validation.
· Week 3: Build the recommendation logic (filter by rating, time control, and error depth) plus spaced‑repetition scheduler.
· Week 4: Record a functional demo, write documentation, and submit.
Why me? I’m a passionate chess player and software developer. I know exactly what it feels like to stare at a board with no plan. This tool isn’t about cheating — it’s about learning, improving, and finally understanding the game at a deeper level.
Thank you for considering ChessMind. I believe it embodies the spirit of Axelera’s challenge: creative, practical, and powered by edge AI.
