what I do best_

Take an app that barely holds up under real traffic and make it reliable.

A lot of products get built fast right now, by a small team or by AI, and they ship. Then they start to creak. Requests time out, the database melts under load, and nobody knows why it broke because there's no observability.

One client brought me a vibe-coded trading prototype that ran out of memory within an hour. I led the rebuild into a system that reliably handles thousands of market-data events a second across hundreds of assets. That's the work I'm best at: find where the prototype stops being a product, then make it hold up without rebuilding more than the business needs.

I build AI products, and the harness that makes them shippable.

Two things, really. The products: RAG systems, evals, fixtures, the agentic workflows and the main chains that do the actual work. And the harness around them: the skills, the adversarial review agents, the MCPs and worktrees that make coding with LLMs produce something you'd actually merge.

I've been living in this for the past year, on my own products and inside an LLM company. It's not magic and I won't tell you it's a 10x. But done right it genuinely produces more, and done wrong it produces garbage code and garbage git hygiene at scale. I know the difference.

Full-stack product engineering, when that's what the job needs.

• Architect and build your MVP

  We figure out what is worth building, then I own the first real version from architecture to launch.

• Own a full-stack product area

  Take responsibility for a piece of your product and ship it front to back, without hand-holding.

• Rescue or refactor an existing product

  Untangle the part everyone's scared to touch, fix the reliability gaps, and make it safe to change again.

• Build data-intensive systems

  Design the pipelines, processing, and dashboards that turn messy, high-volume data into a product people can rely on.

Where the rigor comes from.

Some of what I've built moved real money and data: a game that did $300k in volume its first week, systems handling thousands of events per second, a dApp built to hold $100M+ in TVL. I've also done smart-contract audits and written security research that ended up with ~600 GitHub stars.

I don't chase crypto work anymore, but that world is where I learned what rigor actually costs and when it's worth paying for. If you want the details, they're on the projects page.

A few ways to start.

• Concept sprint · $5,000 fixed

  We whiteboard the idea together and work out what it takes to prove. For a small idea, that might be a rough working demo. For a more complex one, it might be the system design, trade-off research, diagrams, and wireframes, with little or no code. You leave with something concrete and a clear recommendation for what comes next. Not production-ready, and not trying to be.

• Agentic harness setup · $5,000 fixed

  I set up the whole agentic coding harness for your team, tuned to how you actually work. We talk through your stack, your conventions, and what 'done' looks like, then I build the CLAUDE.md files, the skills for your real workflows, the reviewer agents, and the MCP and hooks wiring around them. The value is that it's opinionated & custom and not just a template. If you want to see the shape of it first, my coding-with-agents write-up walks through how mine is wired.

• Weekly-ship subscription · $7,000 per month

  Temporary extra capacity for a busy stretch. We agree on the most valuable slice each week, I own it through review and release, and you get a working increment rather than a pile of hours. Pause or cancel cleanly when you're done. No hire-and-fire, no awkward wind-down.

• Embedded · $800 per day

  Full-time on your product for a few months or longer, the closest to a normal contract. I join the team, own a defined product area, and stay accountable for getting it shipped and running well. For the right team, I'm also open to a long-term engagement or a full-time role.

• A team I stand behind

  If you need more than me, I can assemble and lead a small team of freelancers I've actually worked with and would vouch for by name. I stay your technical point of contact and accountable for the delivery. The value is the vetting, not the headcount.

I work best with founders, CTOs, engineering managers, and product managers at lean, fast-moving startups of roughly 3–50 people, from early stage through Series B. If that sounds like you, tell me what you're dealing with and we'll figure out the right way to start: [email protected].