Lightweight pairwise evaluator for relational signals in Ouro-2.6B-Thinking loop-state trajectories.

A large transformer model trained on 11 million Czech sentences to learn comma correction in text.

A hardware-software co-design of a Mini-Transformer next-token predictor, featuring a Single-Layer Self-Attention pipeline implemented in pure RISC-V assembly and a custom 16-bit processor datapath designed in Logisim with a dedicated ISA optimized for parallel vector dot products and accumulations.

End-to-end PyTorch implementation of a GPT-style decoder-only transformer from scratch. Features custom multi-head causal self-attention, AdamW optimization, and a production-ready inference pipeline with checkpointing.