Services
AI MVP development for startups: a working product with real auth, a real database and real users, polished enough to show investors. Not a Streamlit demo. The code is yours from day one.
Standard 3-week sprint runs $15–22K. Extended (4 weeks with payment, landing page, analytics) runs $22–30K. If you need an AI Product Sprint instead ($30–60K, 4–6 weeks), that's the next tier — more architecture, more polish, more production-grade. Fixed-price per sprint, milestone-based payment.
1-hour call. I write a 1-pager: the single user journey we're validating, what we're explicitly NOT building, the success metric for the validation. Free.
Auth, database schema, primary user journey end-to-end. Ugly but functional. Deployed to a staging URL by Friday.
LLM integration, prompts, basic evals, UX polish. The design background shows up here — your MVP doesn't look vibe-coded.
Analytics, payment (if needed), landing page, launch. Hand off the repo.
Default stack: Next.js + Supabase + LangGraph + Claude. Boring, fast, and the code can evolve into your v1 without a rewrite. I avoid exotic frameworks in MVPs — every day spent learning a new tool is a day you're not learning from users. The AI-assisted development workflow (Cursor + Claude) is what makes the 3-week timeline real; without it, this is a 10-week build.
Real questions from scoping calls. If yours isn't here, ask it on a free 30-minute call.
One user journey end-to-end with auth + LLM integration + polished UX, hosted on a staging URL, source code in your repo. With the 4-week extension: add payment, analytics, landing page. I always under-promise on scope and ship what's in the one-pager, not what creeps in over Slack.
Most founders don't at MVP stage — that's fine. For RAG products, I seed with public data (docs, blogs) or use synthetic data we generate together. For agent products, we focus on workflow logic first; data quality becomes a v1.1 problem.
Yes, from day one. Repo lives in your GitHub org, not mine. Stack is standard and popular, so any full-stack engineer can take it over.
30 days of bug-fix support is included. After that you can either hire a full-time engineer (I'm happy to help interview) or convert to an Agentic Product Retainer for the next phase — most founders pick one or the other after real-user feedback comes in.
Vibe coding (Karpathy 2025) = AI-assisted development where you describe what you want and Claude/Cursor write the code. The technique closed the 'I'm not technical enough to ship' gap for designer-PMs in 2026. My MVPs ship polished because the design + product judgment is the bottleneck, not typing speed.