05

Research workshop

Experiments meant to become working systems.

These are long-running engineering projects, documented while the questions are still alive—not rewritten afterward to make progress look inevitable.

01
SMALL LANGUAGE MODELSACTIVE RESEARCH

Train fluency. Retrieve knowledge.

A compact language model built and trained on consumer hardware. Follow the data preparation, model architecture, and experiments behind a 369-million-parameter model, with search and tool use as future research directions.

  • Consumer-GPU pretraining
  • 32-layer transformer
  • Reviewed data and reproducible training

Explore the project ↗ Training monitor ↗ Try the model ↗

02
SELF-LEARNING CHESSACTIVE BUILD

Learning chess from the board itself.

An AlphaZero-style system with a residual policy-value network, parallel Monte Carlo tree search, and a self-play pipeline built to make serious experimentation possible on available hardware.

  • 1.6M-game source corpus
  • ResNet policy-value model
  • Batched parallel search

Explore the chess project ↗