Hello, I'm
Computer Science graduate specialising in AI/ML and full-stack engineering, and solo founder of RootLayer, an AI-powered web compliance SaaS. I ship production systems end-to-end and bring the same evidence-driven approach to machine learning, forecasting and systems projects.
Learn about my
BSc (Hons) Computer Science
September 2022 – November 2025 | 2:1 Honours
Relevant Coursework: Artificial Intelligence I & II, Machine Learning, Database Design & Implementation, Advanced Software Development, Autonomous Agents & Multi-Agent Systems, Advanced Algorithms, Advanced Artificial Intelligence.
Look into my
Founder
May 2026 – Present
Built and shipped an AI-powered web compliance SaaS solo, covering WCAG 2.2 AA accessibility and GDPR/EAA privacy auditing. Combines an LLM agent driving a real browser with a deterministic axe-core baseline, on full product infrastructure: async job processing, multi-tenancy, white-label reporting, SSRF protection and rate limiting.
Full-stack Engineer Intern
July 2025 – September 2025
Developed a PostgreSQL database with a 10-table star schema storing product emissions, supplier and shopping-history data. Built an ML model analysing 46 nutritional parameters and a recommendation engine that detects nutrient gaps from shopping history.
UX & Accessibility Intern
July 2024 – August 2024
Conducted comprehensive research on AI applications in healthcare, facilitated stakeholder workshops, and developed user personas. Designed service and journey maps, prototypes, and conducted user research to refine solutions. Delivered actionable insights for more user-centric, efficient healthcare AI systems.
Explore My
Python & TypeScript
Full-stack & systems architecture
Backend & API Design
FastAPI, NestJS & REST services
Distributed Systems
Celery, Redis & async worker queues
Relational Databases
PostgreSQL, MySQL & JSONB schemas
High-Performance Systems
JIT compilation, bitboards & algorithmic optimization
Containerization & DevOps
Docker Compose, multi-stage builds & Nginx
Automated Testing & QA
Playwright E2E, integration & unit suites
Version Control & CI/CD
Git workflows, GitHub Actions & release automation
Generative AI & Diffusion
Stable Diffusion, ControlNet & zero-conv pipelines
Multi-Agent Systems
LLM orchestration, tool use & multi-pass adjudication
Deep Learning & PyTorch
Custom training loops, diffusers & CUDA acceleration
Graph Neural Networks (GNN)
PyTorch Geometric, temporal holdouts & embeddings
Computer Vision
Medical segmentation, OpenCV & Canny edge detection
Natural Language Processing
Semantic diffing, filings analysis & embeddings
Autonomous Game AI & Search
Alpha-Beta, PVS, transposition tables & quiescence
Explainable AI (XAI)
Feature attribution, zero-shot validation & interpretability
Probabilistic Modeling
Dixon-Coles & Poisson goal expectation engines
Monte Carlo Simulation
50,000+ run tournament & league distributions
Statistical Machine Learning
LightGBM, XGBoost, scikit-learn & ensemble blending
Walk-Forward Backtesting
Out-of-sample validation & betting market benchmark checks
Data Engineering & ETL
NumPy, Pandas & high-throughput data manipulation
Scientific Visualization
Statistical distributions, interactive metrics & reporting
Specialized Technologies, Frameworks & Tooling
Browse My Recent
Autonomous Web Compliance & Accessibility Platform
AI-powered web compliance SaaS I founded and built solo. Delivers WCAG 2.2 AA accessibility audits and GDPR/CCPA privacy assessments by combining deterministic axe-core and Playwright scanning with LLM adjudication, plus full-site crawling, scheduled monitoring, multi-tenant workspaces and Stripe billing.
NPC Minds for Game Developers
AI toolkit that gives NPCs minds of their own: they remember, hold beliefs that can be false, and act when the player isn't looking. Code decides every action and a language model only voices it, with each line checked against the events and beliefs it cites. In measured runs, 87% of NPC turns needed no model call and the validator blocked 0 of 91 model replies.
GPS-Free London Underground Noise Mapping
Built for Encode London 2026. Turns Tube rides recorded on a phone into measurements placed between stations without GPS, reaching a median of 1.95 m against 8.2 m for a timetable in a pre-registered test on new rides. A Gemini agent drafts a cited evidence pack from TfL records, a deterministic verifier checks every quote and number, and a human approves it.
Colon Cancer Detection Platform
Final-year dissertation: a containerised medical AI platform combining histopathology and colonoscopy video analysis. Pairs an EfficientNet-B4 UNet++ model for real-time polyp segmentation with explainable AI (Grad-CAM, saliency maps), served through FastAPI/PyTorch microservices.
High-Performance AI Chess Engine (AI Chessathon)
Chess engine for the Optiver-sponsored AI Chessathon 2026. Joined on Day 3 and climbed from #241 to #82 of 485+ teams in under 48 hours. Compiling the search with Numba lifted throughput from 13,000 to ~2 million nodes/sec, and changes shipped only on statistically significant results, including a time-management fix worth +61 Elo (95% CI [+35, +88], 513 games).
Probabilistic FIFA World Cup Forecasting Engine
Elo-parameterised Dixon-Coles model fit on ~49,000 international matches since 1872, wrapped in a 50,000-run Monte Carlo bracket simulation. Validated with a leak-free walk-forward backtest against de-vigged bookmaker odds, it ranked the eventual four semifinalists as its top four by champion probability.
Across AI, data science, full-stack and systems work
Parity-Verified Diffusion
From-scratch, mathematical parity-verified ControlNet for Stable Diffusion v1.5 with end-to-end training pipeline, zero-convolution growth validation, and COCO Canny conditioning.
AI 10-K Disclosure Diffing & Removal Detector
AI-powered 10-K risk disclosure analysis engine comparing SEC Item 1A/7A filings across years to surface removed, softened, and new risk language using semantic alignment and LLM judgment.
Pricing a Prop-Firm Challenge
Working paper and full reproduction package pricing a retail prop firm's published evaluation contract on twenty years of intraday index prices at measured broker costs, with every hypothesis fixed and committed before its run.
Macro Alpha from News Sentiment
Pre-registered replication of a paper reporting Sharpe ratios of 4.65 to 5.87 from FinBERT sentiment on GDELT headlines. Result: not replicated. All 18 asset, timing and model combinations fall between -1.00 and +0.02, and none beats random-sign trading at the same costs.
AI Career Advisor
Agentic career advisor that parses CVs, scores each skill against O*NET AI automation exposure data, ranks matched jobs by a fit-vs-risk composite score, generates tailored cover letters, and autonomously submits applications via browser automation with a live CDP screencast feed.
Multi-Entity Command Centre for Xero
Multi-entity financial operations command center where AI proposes structured cross-organization chart of accounts mappings with human approval, deterministic execution, and drift-aware health monitoring.
Get in Touch
