Work
I build AI-native products end to end — from the data and extraction pipelines underneath to the polished front-ends on top — shipped solo or in teams, and I build them fast.
Example Projects
JutJut
jutjut.co.uk →Opportunities for young people shouldn't depend on luck — JutJut connects them to the life-changing opportunities they'd otherwise never hear about.
A youth opportunity discovery platform. A Turborepo monorepo — a Next.js web app, a FastAPI backend, shared design-system and data packages, and Supabase auth — with an AI pipeline that discovers, scrapes and extracts opportunities from across the web, plus a social feed, a personalised feed and admin tooling.
Next.js · React · TypeScript · Tailwind · Turborepo · FastAPI · Pydantic · Supabase
RedLines
In progressCreating easy access to planning applications for the public, industry and government.
A georeferencing engine for planning documents. It recovers the red-line site boundary that exists only as pixels on a map figure inside a PDF and places it in real-world coordinates as GeoJSON — via a cheap-to-expensive ladder (embedded GeoPDF metadata, coordinate-grid affine solve, declared-extent captions, scale-bar-and-anchor, basemap match) backed by PDF parsing, computer vision and OCR. First vertical: UK planning red-line boundaries.
Python · FastAPI · PyMuPDF · GDAL · OpenCV · Tesseract OCR · vision models · shapely · pyproj · GeoJSON
What I do
I build large-scale AI data systems, solo and in teams — turning messy documents and web pages into reliable structured data, and building the checks that make an LLM’s output trustworthy enough to act on. The working method is documented in the Field Notes.
If you have a data problem, if no one else can help, and if the data is out there… maybe you can hire… AukInfo.