4x founding hire. Most recently led the build of a multi-agent AI product with voice, zero to customers in weeks, in a sector that was actively banning AI tools. Before that: a four-sided marketplace in 10 days during a national crisis, and structured workflows in SRA/FCA-regulated legal tech.
Now: Building and deploying AI at Zero Gravity. Multi-agent architecture, voice, correctness evaluation, enterprise partners (HSBC, KPMG, Accenture). Previously: Flash Pack (pre-seed to Series A), MealsForTheNHS (co-founder, Β£1.8m raised, 303k meals), Farewill (SRA/FCA regulated).
Background: Languages BA (Birmingham/Fudan). French & English (native), Spanish (fluent), Chinese (intermediate). BlueDot Impact AI Governance & Alignment.
ποΈ Podcast: Just Now Possible with Teresa Torres. Building AI products that augment human relationships, not replace them.
Domain expertise turned into tools any AI assistant can use. Python, live regulatory data, production-tested.
| Project | What it does | Why it exists |
|---|---|---|
| π clearbook | UK home-buying MCP server β conveyancers and mortgage brokers from SRA/FCA/Companies House registers, stamp duty, lease checks, survey explainer, title register analysis | 11 tools. Live regulatory data cross-referenced with company health. The complete home-buying journey in one server |
| π‘ hiresignal | Open spec + MCP server for AI-queryable professional profiles | Recruiting agents need structured candidate data, not scraped LinkedIn |
| ποΈ vox | Voice of Customer research agent | 8 days of PM research in 8 minutes. JTBD, personas, opportunity mapping from Gong/Granola/Jiminy data |
| Project | What it does | Why it exists |
|---|---|---|
| 𧬠claude-eval-toolkit | Evaluation framework for Claude-powered apps β evals, guardrails, deployment patterns | Test your AI before it reaches users. 69 test cases, LLM-as-judge grading, UK education safeguarding (KCSIE/Prevent) |
| π§ decision-trail | Capture human decisions during AI-assisted development | AI writes the code. Humans make the calls. This records which calls were made and why β so you can audit, learn, and improve |
Tools for AI-assisted development. I use Claude Code daily on a production app β these came from scratching my own itches.
| Project | What it does | Why it exists |
|---|---|---|
| π§ͺ Claude-Skill-Potions | Curated Claude Code skills for ops and product workflows | The skills directory is 40K+ deep. These are the ones that actually work. 51 stars |
| π zeste | macOS menubar app β search, install, and manage Claude Code skills | Finding and installing skills shouldn't require digging through GitHub |
| π claudemd-lint | Linter for CLAUDE.md files | Catches vague rules, bloated configs, and instructions that should be hooks |
| πͺ hooksmith | Browse and install pre-built Claude Code hooks with one command | 12 hooks, zero config. The missing package manager for CC hooks |
| Project | What it does |
|---|---|
| β spawn-cafe | Share your link, pick a slot, find a cafe halfway. Agentic coffee meetup scheduler |
| ε’¨ mo-hanzi | Menubar SRS for learning Chinese characters (HSK 1-3) |
| π plod | A calm running app that cares about your health, not your PBs. PWA with Strava sync |
| π§ spotifyunwrapped | Your Spotify data, visualised properly. Privacy-first analytics, runs entirely in your browser |
- What I Learned Spending A Week With 100+ AI Leaders
- What I Learned Shipping AI to Users Who Can't Afford Bad Advice
- Amsterdam Built the 'Perfect' Ethical AI System. It Still Failed.
- The Trust Gap: Why Your AI Product Will Fail Without This
- Claude Keeps Making the Same Mistakes. So I Started Writing Down the Fixes.
