We measure before we build, and prove it before we ship.
A small practice that builds real software and grounded AI. We start with the numbers, write the thing, and verify it does what we said. No theater, no hand-waving, no demos that fall apart under a real question.
Case studies
Shipped work, told plainly. The problem, the approach, and how we knew it held up.
Demos
Concept builds you can poke at. Small, honest sketches of what a system can do.
How we work
An engine-first method that puts measurement and proof ahead of opinion.
Built carefully, and built to be checked.
A selection of recent case studies. Each one led with a number we could verify, not a promise.
DUSKFALL
A real-estate underwriting platform where one model is the source of truth, and a validator proves the deck, the web proforma, and the spreadsheet can never disagree.
feldstein.travel
A live, hand-built editorial site on Astro and Cloudflare, with its own design system.
Caldrop
A feasibility spike that answered, with real numbers, whether you can parse every US school district academic calendar at scale.
Inbox AI
An email triage system whose language model runs locally on the client's own machine, with a safety scrub gate.
Haus Command Center
A live operations dashboard built over real property-management APIs.
flightscout
A published open-source CLI and MCP server for Google Flights research, with booking-window intelligence and reliable fare watches.
How we work
The engine-first method behind every project. Measure first, build a single source of truth, prove it with a validator, then keep checking after it ships.
Concept builds, made to be poked at.
Small sketches that show the shape of a system. Each one is a concept build, not a client engagement.
Wend Coffee
A brand microsite that shows pure visual craft.
Talk to your business
A grounded retrieval agent that answers from a company's own documents, with citations.
Intake triage bot
An AI pipeline that classifies, routes, and drafts replies to inbound messages.
Automation flow
A live lead pipeline: enrich, score, route, notify.
The engine-first method behind every project.
Measure before you build. It is the one habit that keeps the work honest.
Opinions are cheap. Measured numbers change the conversation, and a validator keeps everyone honest after the meeting ends.
Measure
Pull the real signal first. Volume, cost, feasibility. We ground the plan in what is actually true, not what sounds good.
Build the engine
A single source of truth does the work. Everything the client sees reads from it, so there is one place to be right.
Prove it
A validator checks the output against expectations before anything ships. If the parts can disagree, we treat that as a bug.
Ship, then keep checking
Live work gets a standing check so drift surfaces early. Shipped is the start of the relationship, not the end.