// how models reason — and where they break
Mentored research programme with Stanford/Berkeley researchers. Working on how reinforcement learning shapes chain-of-thought reasoning in large language models, targeting a submittable paper.
Implemented and benchmarked classical vs. quantum models — SVMs with quantum kernels and variational circuits — in Qiskit on simulated backends, exploring hybrid quantum-classical optimisation.
// commit history, newest first
Full-stack work on AI-enabled internal tooling for finance data — retrieval quality, evaluation, and guardrails alongside the application layer. Angular · TypeScript · Scala · Python · SQL.
Secured £50,000 in industry sponsorship for Durham's 600+ participant hackathon and managed relationships across corporate sponsors.
Financial markets and fintech infrastructure through workshops and team projects across technology and markets divisions.
Automated client-communication workflows, cutting projected print and postage costs by 50% and ~12,000 pages of paper a year.
Built a PDF→CSV extraction tool saving ~£12,000 a year; developed and deployed a multi-site website supporting 30 practices.
// built fast, shipped anyway
Mental-health AI application built in 24 hours: facial recognition to engage users, movement tracking feeding a personality quiz, and an analysis engine generating personalised responses.
Voice-first real-estate lead engine: a buyer talks to a voice agent for 90 seconds and gets a WhatsApp with matched London listings plus a scheduled callback — no human in the loop. Built the n8n orchestration: trigger → rank → notify → log back to Attio.
AI matching engine scoring company fit from a candidate's GitHub, CV and personal statement via GPT-4o semantic reasoning. Full-stack with GitHub OAuth, deployed on Vercel.