Promptomize.
A supervised rewriter that improves accuracy on any frozen LLM backend, with a multi-backend evaluation harness across GPT, Claude, and Gemini.
Read full case →
A clear point of view ships faster and ages better than consensus. We argue an idea hard, then commit and build it.
A live product is worth more than a portfolio piece, every time. We measure quality by what's deployed, not what's framed.
SwiftUI for iOS. Real React for web. Server-rendered when it should be. The shortcut you take in week one is a tax you pay in year three.
We use LLMs in our pipeline daily. We rarely sell them as the headline. The interesting product is the one where the AI is invisible.
A small portfolio of work per year, six-week minimum per engagement. Depth comes from rarely being forced to context-switch.
If a thing is hard, we say it's hard. If a deadline is at risk, you'll know before the deadline. Surprises are the opposite of professionalism.
A supervised rewriter that improves accuracy on any frozen LLM backend, with a multi-backend evaluation harness across GPT, Claude, and Gemini.
Read full case →
Point a phone at a room. Get an AI-rendered remodel and a line-itemed estimate — image generation by Google's Nano Banana via the Gemini API.
Read full case →
A focused budgeting app for iPhone. Spend tracking, category goals, and recurring-charge detection — built to make the next financial decision easier, not heavier.
Read full case →An AI interview coach that runs realistic mock interviews over voice, scores each answer against a per-role rubric, and turns the recording into a study plan.
Read full case →A vehicle intelligence platform with an audio-diagnostic mode: record what your engine, transmission, or wheel sounds like, and an AI classifier surfaces the most likely cause before the mechanic does.
Read full case →SwiftUI, App Store launch, in-app purchases, analytics, support flows. Production craft in a six- to ten-week build cycle.
React + Next.js, server-rendered, accessible, deployed on Vercel or Cloudflare. Marketing sites that load in under a second.
RAG pipelines, prompt optimization, agent harnesses, custom fine-tunes. Working demo in two weeks; production-hardened in six.
A three-day discovery: the problem, the user, the constraints. Outputs are a one-page brief, a paper sketch of the flow, and an honest go/no-go.
No Figma marathons. A working prototype on real devices by end of week two. Wrong faster, learn cheaper, iterate sharper.
Three weeks of focused build. Daily deployable trunk. Friday demo to you. Production-grade tests, observability, and rollback paths from day one.
Documentation written for the next engineer, not for archive. Two weeks of post-launch support included. Retainer for ongoing work, optional.
Two paragraphs is plenty. We respond within one business day, in plain language, with an honest assessment of fit.