blog18 posts3 categories12 active 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.

Start here3 posts that show the work

Building SoberMark: twenty public datasets, one record per provider
SoberMark is a national addiction-treatment directory I build and run solo under my LLC. It assembles 10,900+ providers from twenty federal and state data sources, matches the same facility across sources into one record, and serves it through PostGIS and full-text search. I built it to get deep on databases: search, cleaning, combining, and managing data in production.
postgres
postgis
search
data-engineering
python
nextjs
product
October 6, 20263 min
Building SoloSteady: applied AI where the numbers have to be right
SoloSteady turns twelve months of bank deposits into an income summary for self-employed people applying for a bank-statement mortgage. I build and run it solo under my LLC. I set out to learn applied AI and agentic systems in production. The biggest lesson was where an AI model belongs in a money document, and where it does not.
applied-ai
evals
fintech
plaid
nextjs
spec-driven
product
October 6, 20263 min
Loom: durable AI commerce with Vercel Workflows, exactly-once side effects, and a bounded agent gate
Loom is an open-source backend that demonstrates the patterns required to let an LLM influence real money movement safely. Four Vercel Workflows run end-to-end with durable sleep, exactly-once side effects, saga compensation, Stripe webhook drift reconciliation, and a deterministic authorization gate that bounds every agent decision. Ships with an adversarial eval harness and a failure-injection harness. Built on Vercel Workflows (GA), the Vercel AI SDK, Anthropic, Stripe, and Upstash.
ai
durable-execution
agents
workflows
stripe
open-source
applied-ai
May 23, 202614 min

series / building in the open / 10 posts

20264 posts

Notes from the field.

Building SoberMark: twenty public datasets, one record per provider

Building SoloSteady: applied AI where the numbers have to be right

Loom: durable AI commerce with Vercel Workflows, exactly-once side effects, and a bounded agent gate

Building issuegraph: a LangGraph triage agent, with the evals to prove it works

Building anchor: dual-surface product pages, agents.json, delegated-authority checkout, and AEO instrumentation

Building Forge: a multi-agent debugging concierge in production-grade TypeScript

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

Building grant-pilot: a multi-turn agent that orchestrates three sub-agents over real federal data

Building fedbench: an LLM eval harness for grounded Q&A

Building fieldops-mcp: designing the tool surface an agent actually uses

Auditing my own portfolio against Next.js 16: what I shipped, what I skipped, and why

Frontend is a discipline, not a stylesheet

What five years from bootcamp to senior taught me

Why I'm not chasing FAANG (and where I think frontend talent is actually moving)