Model evaluation
Urban Jungle had no reliable way to tell how well its models were performing. I'm building benchmarks with DSPy and GEPA so every change to a model can be measured against them.
- Python
- DSPy
- GEPA
Senior Software Engineer in London.
01 / About
Full-stack engineer with strong front-end craft and an eye for design.
I'm a full-stack engineer at Urban Jungle, an insurtech. I'm tech lead on Non-Standard Home insurance, a new product, and I work with the company's AI team on how its models are evaluated. I have three years of professional experience in TypeScript and Python across CGI, a start-up and Urban Jungle.
Outside work I co-founded UniSoc, a society management app, and built Doorstep, a job outreach tool. I have a first-class Computer Science degree from the University of Exeter.
02 / Work
Co-founder
An all-in-one app for running university societies, built from scratch by a team of four. I designed the interface and interaction patterns with my co-founders. Live on the App Store and Play Store, with pilots running at two societies. I also built a UniSoc MCP server that lets Claude read society data such as events, members, bookings and analytics.
A Chrome extension that fills in job applications on any site from your own knowledge base, built from your CV and a mock-interview onboarding. Open questions are drafted from your own stories and marked for review, and sensitive questions and consents are always left to you. Built with a team of four engineers in London, now in pilot with a waitlist.
Solo build
Finds small businesses through the Google Maps API, summarises their websites with an LLM, and drafts emails you send from your own inbox. The first person to use it emailed around 80 London studios and got about 10 replies and a job.
Urban Jungle had no reliable way to tell how well its models were performing. I'm building benchmarks with DSPy and GEPA so every change to a model can be measured against them.
My dissertation: a convolutional neural network that classifies star clusters from telescope imagery, from data preprocessing and augmentation through to model design and evaluation.
An audio deepfake detection tool using CNNs to spot manipulated media.
95%accuracy, real vs fake audio
03 / Experience
Aug 2025 — Present LondonCurrent
Tech lead on Non-Standard Home insurance, a new product, and work with the AI team on how the company's models are evaluated.
Sep 2024 — Jun 2025 Remote
Lead developer on a full-stack desktop app at a start-up, from first architecture to release.
Jul 2023 — Jul 2024 Leatherhead
A year-long industrial placement on web services for a major defence-sector client.
2021 — 2025 Exeter, UK
Including a year-long industrial placement at CGI.