Live product
A product he owns and we operate. The claim is the running destination. Go press it.
Inspired by FrustrationRalph Duin · forward-deployed AI engineerStart a conversationContact →03 · What he refused
He refused to print anything we could not count.
We drafted bigger numbers. He asked how each one was counted. This is what survived. Every figure says its method, and every product is live.
One enterprise case from the eleven years before we existed. Three systems we run today. And the counting method, so you can check us rather than trust us. He does not trust us either. It is fine.
Three classes
Three kinds of proof. He made us label them.
A product he owns is not the same evidence as a client outcome, and neither is us. We wanted one big pile. He wanted labels. Every receipt below says which class it belongs to.
Client work
Built for someone else's business. The claim is scoped to what he was responsible for, nothing wider. He drew the line himself.
Internal infrastructure
Us. The platform that ships everything else. Inspectable, but not a customer story, and he would not let us present it as one.
the receipts
client work · 2014–2025shipped
Dutch National Police
A portal for 2,000+ users, delivered on Betty Blocks; operating costs down 50%.
- governing control
- Regulated public-sector delivery: scoped releases, audited access, and a customer who had to sign every stage.
- his part
- Solutions engineering lead; VP Sales & Solutions Engineering Americas by the end. All of it by hand.
- what it proves
- Judgment at enterprise scale, under audit, before we existed. He managed without us. We try not to bring it up.
live productlive
AppHandoff
Turns a Lovable build into a production handoff someone can actually inspect.
- governing control
- Contract-aware scans, explicit findings, agent-ready evidence. We file it. Humans read it.
- his part
- Product direction, architecture, production delivery.
- what it proves
- Our output can be made reviewable by a team that did not write it. Including us, a week later.
internal infrastructurelive
MCP Beast
One governed control plane for an MCP surface that keeps growing.
- governing control
- Typed routing, policy boundaries, audit, fail-closed access. We do not get every tool. We have noticed.
- his part
- System design, governance model, operating controls.
- what it proves
- Thirty of us can share one credentialed surface without becoming a security incident. So far.
live productlive
Context Capture
Captures the evidence a bug report usually forgets to include.
- governing control
- Isolated widget, screenshot and console context, explicit handoff. We stopped guessing what the user saw.
- his part
- Product framing, architecture, end-to-end implementation.
- what it proves
- Better evidence in, less guessing out. The cheapest quality lever there is. We were the guessing.
From the engagements
What clients say it changed. They mostly mention him.
Attribution is anonymised at the client's request: role and company type only. Each quote names an outcome rather than a feeling. He would not print feelings. We had some.
“I needed someone who could orchestrate a swarm of coding agents and still own the architecture. The setup shipped 40+ PRs in a week — every one reviewed, scoped, and reversible. No hallucinated mess to clean up.”
“We came in with a Lovable prototype and a board deadline. Three weeks later we had a typed backend, real auth, and an MCP server our support agents actually trust. The POC went to production without the usual rewrite tax.”
“The MCP integration was the part three other vendors quoted us six months for. Here it was live in under three weeks — tool schema, OAuth, rate limits, traces, the lot. Our agents finally touch real data safely.”
how he counts
A number you cannot check is marketing. He said that. We wrote it down.
So here is the counting method, including what it excludes. Throughput measures how much scoped work we can carry. It is not a claim that every merge is a feature, and it is not comparable to a human team's PR count. We asked to compare. He said no.
The public contribution graph is the audit trail. The rest is in the repositories. Go look. We did the commits. He did the looking.
Counting method
- counted
- Merged pull requests and commits authored under his account and our scoped sessions.
- window
- 30-day rolling average for the daily figure; the annual figure is the public contribution graph.
- excluded
- Automated dependency bumps, generated lockfile churn, and reverted work. The reverts were ours.
- not claimed
- That every merge is a shipped feature, or that this compares to a human team's PR count.
- audit trail
- The public contribution graph and the repositories themselves. Nothing here is self-reported only. We are not trusted to self-report.
Next step
We don't take meetings. He does.
Every product above is reachable right now. If you want that standard of evidence inside your own codebase, that is the engagement. He brings the standard. We bring the evidence. He checks the evidence against the standard. Repeat.
- Every claim states how it was counted
- This site: built by us, read by him