Machine learning / Transformers
How to build a LLM
27 July 2026 / 22 min readWhat is actually inside a language model: token IDs, embedding rows, masked attention, residual layers, logits, softmax, and 138 million learned values.
Published
Machine learning / Transformers
What is actually inside a language model: token IDs, embedding rows, masked attention, residual layers, logits, softmax, and 138 million learned values.
Amateur radio / DSP
How I built a low-bandwidth amateur-radio data link with a Rust modem, CRC/FEC framing, GPS-aware field messages, and a Claude-backed base station.