Blog
- Building SoberMark: twenty public datasets, one record per provider
SoberMark is a national addiction-treatment directory I build and run solo under my LLC: 10,900+ providers from twenty public data sources.
- Building SoloSteady: applied AI where the numbers have to be right
SoloSteady turns a year of bank deposits into an income summary for self-employed mortgage applicants. Where an AI model belongs in a money document.
- Building issuegraph: a LangGraph triage agent, with the evals to prove it works
An AI agent that sorts GitHub issues and drafts replies, built on LangGraph with human review steps and evals that check its confidence.
- Building anchor: product pages for humans and AI agents, agents.json, delegated-authority checkout, and AI crawler analytics
An open-source product catalog built for people and AI agents: agent-readable pages, a signed agent checkout, and analytics for AI crawlers.
- Loom: AI commerce on Vercel Workflows with durable execution, exactly-once side effects, and a bounded agent gate
An open-source backend that lets an AI agent take part in real payments safely: exactly-once side effects and a hard limit on what it can approve.
- Building Forge: a multi-agent debugging assistant in production-grade TypeScript
An open-source debugging assistant: four AI agents investigate a stack trace in parallel and a coordinator ranks their likely causes.
- Auditing my own portfolio against Next.js 16: what I shipped, what I skipped, and why
I audited my own portfolio against the Next.js 16 upgrade guide line by line: what the site uses, what it leaves out, and what I caught.
- Building Kev-O: a RAG chatbot that only answers from my own writing
Kev-O answers questions about my work using only what I've published here. How it finds answers, cites sources, and keeps the API bill capped.
- mdx-corpus: a tiny package that turns an MDX directory into retrieval-ready chunks
A small npm package that turns blog posts into a search index for AI chatbots. I pulled it out of Kev-O: files in, JSON out.
- streamfield: a small library for AI streams that don't look broken
When the Vercel AI SDK streams a structured response, fields flicker into place. streamfield is a tiny React library that fixes that.
- Building tablesalt: a CSV agent where the answer IS the UI
An in-browser data agent: drop in a CSV, ask a question, and the answer comes back as a chart, stat, table, or list. No chat bubbles.
- Building grant-pilot: a multi-turn agent that coordinates three sub-agents over real federal data
An AI agent that finds federal grants for small businesses and nonprofits, with three sub-agents, a hard daily budget cap, and a planner that never throws.
- Building fedbench: an LLM eval harness for document Q&A
fedbench tests how accurately an AI answers from policy documents. Most document agents fail in three ways, and it measures all three.
- Building fieldops-mcp: designing the tools an agent actually uses
An AI assistant that schedules and dispatches field-service jobs through a small MCP server. The hard part was deciding what it can do at all.
- Frontend is a discipline, not a stylesheet
Engineers who call frontend trivial are often the ones rerouting every CSS ticket. What 'just CSS' actually costs to ship.
- What five years from bootcamp to senior taught me
Seven patterns I noticed somewhere between bootcamp in 2018 and senior at Deloitte around 2023, written for the version of me at year two.
- Why I'm not chasing FAANG (and where I think frontend talent is actually moving)
The default career advice for senior engineers in 2026 is wrong for at least half the people getting it, and the reasons have shifted.
- An anthropology degree was the most useful thing I brought to engineering
The major engineers laugh at quietly turned out to be the edge, and the branch of anthropology I picked is the one with the most well-trodden path into tech.