blog11 posts30 topics

Notes from the field.

Working notes on the craft — engineering at scale, the AI-era developer experience, frontend that most teams leave on the table, and whatever else has my attention this week.

Building Kev-O: a grounded RAG chatbot trained on my own writing
Kev-O answers questions about my work using only the public corpus I've written. Hybrid BM25 plus cross-encoder rerank, Claude streaming through the Vercel AI SDK, a daily USD cap so it can't blow up my API bill. Three surfaces, one brain. Here's what I built and what choices ended up mattering.
ai
rag
retrieval
streaming
open-source
applied-ai
May 15, 20268 min
mdx-corpus: a tiny package that turns an MDX directory into a retrieval-ready corpus
While building Kev-O I kept rewriting the same parse-and-chunk step for MDX content. So I pulled it out into a small npm package that does file-in / JSON-out and gets out of the way. No embeddings, no vector store, no LLM. Three responsibilities, sharp boundaries.
ai
rag
open-source
npm
mdx
corpus
May 15, 20266 min
streamfield: a small library for AI streams that don't look broken
When the Vercel AI SDK streams a structured response, the fields flicker and snap into place as they arrive. streamfield is a tiny React library that fixes that. One component, four props, no dependencies.
ai
react
vercel-ai-sdk
open-source
frontend
May 14, 20264 min
Building tablesalt: a CSV agent where the answer IS the UI
tablesalt is an in-browser data agent. You drop a CSV, ask a question, and the agent renders the answer as the right kind of UI: a chart, a stat card, a table, or a list. No chat bubbles.
ai
generative-ui
vercel-ai-sdk
vercel-ai-gateway
evals
open-source
May 14, 20263 min

Building Kev-O: a grounded RAG chatbot trained on my own writing

mdx-corpus: a tiny package that turns an MDX directory into a retrieval-ready corpus

streamfield: a small library for AI streams that don't look broken

Building tablesalt: a CSV agent where the answer IS the UI