Building
The home page says it: copy-paste programmer by night. I have written code since high school, I am not a professional engineer, and I have never let that stop me from shipping things. Today most of what I build I build with AI coding agents, which has made the gap between “I can describe it” and “it runs” smaller than it has ever been. This page is what I have built and what I am building now.
The intelligence platform behind RTP
The biggest thing I have built is not a side project. When I joined RTP Global in 2021, I started the firm’s data and intelligence capability from scratch, with the first engineers joining shortly after. The result is an in-house data insights platform that tracks hundreds of thousands of companies and close to a million people on hundreds of data features, surfaces signals to the deal team, and shows up where the team already works: a web and mobile interface, a Slack bot that answers in deal flow channels, and a private Chrome extension that annotates a startup’s website, a founder’s LinkedIn profile or a GitHub repository as you browse. On top of it runs a machine-learning model trained on past venture vintages that ranks the companies the platform flags.
We chose to build rather than buy, so that we own the data and the model. The reasoning and the architecture are in Why We Believe in Data Driven Venture Capital. The team that builds it with me is small, five people when I wrote that post, and I still spend a good part of my week in it.
The 30-day AI sprint
In April 2026 I pulled all 30 people at the firm, partners, investors, finance, legal, EAs and ops, out of business-as-usual for a month to build with AI. To make that safe I built the platform layer first: an internal hosting layer on Cloudflare Workers for Platforms with isolated execution per app, single sign-on in front of everything, secret management, and automated checks on every pull request, because the bottleneck with fifty first-time builders is review. By Demo Day the firm had shipped 52 internal apps and several production agents that handle real work end to end. The full story, including what broke, is in Building an AI-native firm in 30 days. Except we didn’t.
Small things, for fun
- This website. A Hugo site on Cloudflare Workers, with a hardened content security policy, a Person-centred schema.org graph, and plain-text and Markdown versions of every page for language models and agents (see llms.txt). The dotted globe on the home page is a few kilobytes of dependency-free canvas code: a Fibonacci lattice, a land bitset from Natural Earth, the real day and night terminator, and a guess at where you are from the Cloudflare data centre that served you the page.
- Scrapers and data tools. Most of my OSINT work ends up as small scripts that collect, clean and join public data. The useful ones grow into pipelines inside the platform above; the rest live on my laptop until the next question.
- Agents. Like everyone in the sprint, I run a personal agent on a hardened VPS that knows my calendar and talks to me over Telegram, and I keep adding to it. Half the value is in what it does; the other half is in understanding, as an investor, what the AI infrastructure companies I meet are actually building.
How I build
Describe the outcome, let an agent write the first version, read every line, test it, and keep only what I understand. Ship small, ship often, and prefer boring infrastructure. Everything public is on GitHub. If you are building developer tools, AI infrastructure or anything that makes people like me more dangerous, I would like to hear from you: that is what I invest in.