// Course · Generative and agentic AI
A chess language model, from scratch.
Thirteen modules building a single project: a GPT that learns to play by reading games, gets aligned with rewards a chess engine can verify, runs in your browser and ends up as a coaching agent with tools. The lessons are written in Spanish.
A closed domain, open data, small models.
- 01
Verifiable
Move legality and quality are decided by an engine, not a person. That makes reward models, DPO and GRPO possible with real, cheap rewards.
- 02
Open
Lichess publishes billions of games, puzzles and evaluations under CC0. Nothing to negotiate, nothing to hide about where the data comes from.
- 03
Small
Tens of millions of parameters: trains in hours on a single GPU and fits in the browser. Everything can be measured and published.
Rukh, on one page.
- borja-glez/rukh
Model, data, training, evaluation and agents. Python 3.12, PyTorch, MLflow.
- borja-glez/rukh-lab
This site: the course, in Spanish. Static Astro with Preact islands.
- borja-glez/rukh-web
The demo: play against the model in the browser with onnxruntime-web.