Writings

A decision model picks from your list of answers instead of writing a reply. Built up from how an LLM works to reading answer probabilities from GPT-2 and Qwen2.5, a trained head, and the internals of Laya, an open decision model.

A blog series about building AI agents from first principles. An LLM API is stateless: every turn you send the whole conversation as a JSON array and get text back. A harness is the program that builds, maintains, and protects that array; everything the field calls agents is a set of patches to that one loop.

How we built WSO2 Integrator Copilot, a domain-specific coding agent with 23 tools, 4 subagents, an on-demand knowledge graph, and a caching strategy that hits 81-90% prompt cache reuse.