RESEARCH TO PRODUCTION (WEBSITE W.I.P. - THE AI SLOP WILL BE REPLACED SOON)
I build LLM-powered products and care that they are correct. I studied computer science at the University of Edinburgh and left at 20 to build full time.
Took an AI product to real revenue as a founding engineer at 19
01 · The Arc
Started computer science at Edinburgh at 16. Spent a year on theorem provers, concurrency proofs with Rob van Glabbeek and the IEEE-754 standard in Lean, then took a year out as a founding engineer and helped grow an AI product to real revenue.
Left at 20 to build full time. Now exploring robotics, world models, and infrastructure for AI, and advising a fund. The same instinct the whole way through: make it provably true, then make it ship.
still mostly interested in things that are provably true.
02 · Now
Most AI fails quietly, in ways no one measured. I'm building toward evaluation and reliability infrastructure for the next wave of systems, robotics, world models, and AI-driven infra, where being wrong has a cost. Using the same approaches I learnt in past research, but now looking forward into markets and technology that barely exist. Contact me if you're building here, or just want to chat about it.
Focus areas
03 · Selected work
2025 · Founding engineer
LLM-powered SEO that surfaces brands inside chatbot results. Team of three. I owned the product end to end, embedded with enterprise customers including billion dollar companies, scoped requirements, and shipped fixes the same day from live feedback.
Grew it to meaningful recurring revenue · backed by the CEO of Perplexity and the founding CTO of Slack
President
President of one of the UK's largest computer science societies. Ran HackTheBurgh for hundreds of hackers and a committee of 25, and was invited to meet Princess Anne in recognition of the work.
Major sponsorship secured · Meta, Jane Street, etc.
2026 · Completed
A durable memory layer for Claude Code. It indexes prior session transcripts and injects retrieved context into each turn through hooks. Retrieval runs over two stores, raw conversation and structured claims, served by a sidecar daemon at low latency.
Cited by Garry Tan, CEO of Y Combinator.
Evaluation vs baselines
2023
Contributed LLVM lowering, including the initial GEP implementation and refactors, to a research compiler framework.
523 stars on GitHub
2023 to 2024
Formalising the IEEE-754 floating point standard in the theorem prover Lean with university researchers, proving the rounding theorems. Based on the Rocq (formerly Coq) formalisation of floating point arithmetic.
04 · Get in touch