What does Brandon Crenshaw build?

AI products and the agents behind them: LLM gateways, voice agents, retrieval systems, and the pipelines that run a business around them. Eleven products are live and linked on the work page, with case studies covering how the hardest ones were built.

What does it cost to work with him?

Project work starts at $5,000 a month for a single seat covering product, engineering and go to market. Shorter advisory engagements are priced per engagement. Pricing is on the site rather than behind a call.

Is he available for a full time role?

Yes. He is open to senior and founding AI engineer roles, remote or hybrid in Chicago, and would relocate to San Francisco for the right one. The contact form has an option for exactly that.

TypeScript and Python across the products, React and Expo on the front end, Postgres and Supabase for data, and Claude, Gemini and OpenAI behind a gateway that reports the cost of every call. Infrastructure runs on Vercel, Railway and Google Cloud.

How does he keep AI systems from making things up?

Scoring and decisions run in code, and the model only explains what the code computed. Every number on a page traces to a source, citations are checked against the evidence the model was given, and anything a model proposes that costs money needs a person to confirm it.

How fast can something ship?

A working prototype in one to two weeks, and a production system with tests, monitoring and a rollback path in four to eight. Eleven shipped products and a public record of what each one does and does not prove is the evidence for that pace.

Work — Brandon Crenshaw

Work — Brandon CrenshawEvery product and system I have shipped, live and linked, plus the mobile apps in progress.

Lobstack: A receipt for every model call — An OpenAI-compatible gateway whose real output is not the routing. It is the line underneath: which model served the call, what it cost, and what any saving was measured against.

Circuit: The conversation is the canvas — An MCP server that turns a sentence into a workflow board and then runs it one deterministic step at a time — on the connectors you already gave Claude, holding no credentials of its own.

GLOWUP: The app holds no secrets — A consumer photo-analysis app where the phone never sees a model key, every result passes a schema and hard bounds before it is saved, and minors never get a generated image.

ROAM: The model never makes up a number — Travel ranking where deterministic engines compute every score and fact, and the language model may only explain them, with its citations checked against the evidence it was given.

Otto: Four checks before a stranger answers — A phone agent where the AI disclosure is built in code and verified three times, the number gate fails closed, and the thing it is least allowed to do is spend your money.

This site: A page that cannot lie to you — The rebuild of brandononchain.com: unsupported claims deleted, status computed daily instead of typed, weight enforced in CI, and every demo labelled with what it is really showing.

Arcis: A treasury an agent can use — Seven contracts giving autonomous agents somewhere to hold idle USDC, credit priced off an identity score, and a keeper that harvests, guards the reserve and watches its own gas.

Sentric: One answer, a tenth of a cent, no account — Solana wallet activity turned into a scored signal an agent can buy per call over x402, with the conviction weights published rather than hidden.

Outreach: Where the address came from decided everything — A campaign that learned the hard way that a plausible address is not a contact, and the guardrails that came out of it. Figures refresh from the send log.

swarm-gcp: The same pipeline, on managed pieces — The lead swarm ported to Cloud Run jobs, BigQuery and Pub/Sub in Terraform, with one identity per component and a walkthrough that teaches the stack. Built, not yet deployed.

Jobhunter: Retrieval was the easy half — A job-search agent that scored roles with hybrid retrieval, then learned the expensive way that a plausible email address is not a contact. One in four inferred addresses delivered; published ones, nine in ten.