About

KeyVec is the personal site of keyvec: part introduction, part blog, and part project showcase. The current center of gravity is LLM systems and the practical work behind data pipelines, retrieval systems, agent traces, data quality, and evaluation loops.

The name points at keys and vectors: identifiers, embeddings, metadata, traces, and the small decisions that decide whether a model system can be inspected, debugged, and improved. Posts here favor durable arguments and implementation detail over announcements.

Core Threads

• Writing about LLM systems, data, and evaluation
• Building small tools around retrieval, agents, and data quality
• Sharing what the work turns up, in public

What I Write Down

Essays, project logs, corpus design choices, filtering heuristics, retrieval evaluation failures, annotation schemas, agent memory records, and the operational details that make LLM systems reliable enough to ship. The unifying question: what evidence should technical work leave behind?

Elsewhere

Public code and small tools live on GitHub. Writing flows through the RSS feed, and every change to the site is logged in Changes. Corrections and pointers are welcome; open an issue on GitHub.