Final-year Computer Science student at The University of Manchester. I build AI systems, trading tools and products people actually use. I founded FutureFlow Tuition, co-founded Lumara, and passed the FTMO evaluation to trade a funded account.
An interactive web simulator of the Enigma cipher machine. Instead of only mapping letters, it models the electromechanical path: current flows from a key through the plugboard, three rotors and the reflector, then back out to a lamp, and you can watch each hop as it happens.
IBM Z Datathon 2026Demo coming soonProblem debt shows up in bank data weeks before someone misses rent: a falling balance, payday loans, bounced bills. This system spots it early. An XGBoost model scores each customer's risk every week and explains why they were flagged, what-if scenarios show which changes would lower the risk, and IBM Granite running locally writes a supportive action plan with safety rules so it never suggests more borrowing and always points to free help. Everything runs on IBM LinuxONE, so financial data never leaves the machine.
Nebius × NVIDIA · Physical AIDemo coming soonA cheap scout robot that goes into a building before firefighters do, finds people and reports where they are. The prototype is a toy RC car with one FPV camera and a £5 ultrasonic sensor, with no LiDAR and no map. NVIDIA Nemotron vision models on Nebius Token Factory read the camera and choose each move, a planner surveys the room and heads for open doors, and a ground-plane calibration turns the single camera into a distance sensor.
It found the casualty even when starting facing the wrong way, and left the room through an open door to spot a target in the hallway.
OxHack 2026Demo coming soonA Mafia game played over a call against AI townsfolk who watch your face for tells. A lie-detection model trained on a published deception dataset feeds a Bayesian town: each AI player keeps a probability that each person is the mafia, trusts the detector only as far as it has proven itself on you, and votes to maximise the chance the mafia goes. Every character has its own ElevenLabs v4 voice with delivery cues, so the narrator whispers at night and players get sarcastic when they don't believe you.
A placebo test with tell-free players caught the AI finding signal in noise and the detector leaking players' roles mid-game. Both are fixed, and a test fails if the leak returns.
Research paperDemo coming soonA Python backtester for 15-minute SPY data that works through three strategy families: moving-average trend filters, ICT Judas Swing liquidity sweeps and Fair Value Gap entries. Every trade pays a 0.01% cost, and results are stress-tested with 1,000-path Monte Carlo resampling. The best variant buys the first touch of a Fair Value Gap in the 09:30–11:00 session.
60-day test window, 1,560 fifteen-minute bars.
A free, lightweight trading journal I built when I couldn't find one that wasn't clunky. Log a trade with entry, exit and notes, and it works out P/L and keeps a running win rate. You can attach chart screenshots to each trade and sketch setups in a separate notepad, all beside a live TradingView chart. Everything is saved in the browser, so there's no account and no server.
My first plan was a custom live chart. API rate limits, unreliable WebSockets and iframes that block screenshots sent me back to a simpler design that works.
Team projectDemo coming soonA meal planner that suggests recipes from what is already in your pantry, filters out your allergens and dietary preferences, and lays the week out on a calendar. Missing ingredients go straight onto a shopping list. I built the front-end pages in React and TypeScript.
FounderDemo coming soonAn online 1-on-1 tutoring and mentorship business. I built the site and the student experience behind it: dashboards, per-subject progress trackers, Notion student portals and automated onboarding.
Co-founderDemo coming soonA studio building full-stack web applications and AI integrations for clients. We also built the Lumara Web Dev Academy, the platform behind our work-experience placement: a two-week course in HTML, CSS, JavaScript and Git across ten modules, with four small projects and a capstone reviewed by the team. Students track progress against a 58-item checklist, mentors and reviewers get their own areas, and each student finishes by building a page on a live client site.
Published on RobloxDemo coming soonA social-deduction game set in a 1930s speakeasy. Each player is secretly dealt a role (Mafia, Doctor, Detective, Jester or Villager), then everyone talks, bluffs and votes over voice chat to find the Mafia before they take the town. At night the Mafia can whisper to each other, and the dead only hear the dead. Hosts can run private parties with a code and set the timers, role counts and Jester, or players can jump into Quick Play. I keep shipping updates, most recently the Jester role.
Online one-to-one GCSE and A-level tuition and mentorship. I built the site and the student experience behind it: a booking flow, student dashboards, per-subject progress trackers, Notion student portals and automated onboarding.
Full-stack web and AI studio. Set up CI/CD with GitHub Actions and Vercel, applied LLMs with structured knowledge bases for explainable insights across trading, sports and automotive projects, and launched a multi-tenant trading journal platform with authentication and real-time analytics (Next.js, TypeScript, PostgreSQL, Prisma, Redis).
Helping neurodivergent and autistic students get strong grades in GCSE and A-level Maths, adapting how each topic is taught to the way they learn.
Online one-to-one GCSE Mathematics tutoring.
24 hours, team of four. Built an AI early-warning system on IBM LinuxONE that flagged 98.6% of unseen customers before they missed rent, with a median 11 weeks' notice.
Built Tilt, a Mafia game with an AI lie detector and Bayesian AI players, competing for Hudson River Trading's data and predictions prize and MLH's Best Use of ElevenLabs.
Passed both stages of the FTMO evaluation (Challenge and Verification) within strict drawdown limits and profit targets, then traded a six-figure funded account across indices and FX with fixed stop-losses and position sizing. Earned $5,215.34 in trading rewards by May 2026.
London. Built an agentic retrieval-augmented generation (RAG) system and trained a forex time-series prediction model in TensorFlow on tick and historical market data.
London. AI-focused business proposals and market research.
Built an AI tool that compares UK LiDAR and Sentinel satellite data over time to find coastal erosion and flood-risk hotspots, and presented it to the project team.
Server fail-log analysis in Python and pandas.
Taught primary through A-level Further Maths and GCSE Physics.
Wrote code for DNA sequencing analysis, working closely with the Biology department on a subject that was new to me.
Remote, during lockdown. Co-built a Flutter app that predicts how a person might age from a photo, and pitched the prototype and a business case to a company.