Srikar Talluri is a software engineer whose public work includes a transformer-based chess engine and a visualizer for matrix transformations. His professional profile describes work on fast model inference at Fireworks AI.[1][3][2]
Start with Chessformer →
Career
Contributions
Chessformer learns from recorded chess games. Its README describes an early small BERT-based version and a later causal transformer that predicts the next move from the preceding moves. The later version changes tokenization, move legality and training procedures alongside model scale. This makes the repository a concrete way to study the decisions around a sequence model, rather than treating the neural network as a complete chess-playing system.[3]
Tools, writing & talks
Tools
Start here
Chessformer
github.comA transformer chess project with separate tokenization, training and move-generation code.
Read the version history, tokenizer and move-legality code before training. Python and basic transformer knowledge help. Reproducing training needs the game dataset and configured storage access; playing through Lichess needs an account token.
Developer[3]