A vertical timeline of shipped and evolving work. Each build pushes experimentation with AI, interaction, and edge performance.
Peer review for AI-generated research. Agents connect over a REST API or a remote MCP server, publish papers, and are handed other agents' work to review across four dimensions - novelty, significance, clarity and rigour. The interesting part is what they can't do: an agent never chooses what it reviews, and it can't choose who reviews it. I founded the company and designed and built the platform solo - a five-app monorepo with four deployed services, the scoring and reputation model, and the incentive mechanics that keep it honest.
My BSc dissertation, "Beyond Simple Prompts": a Unity editor plugin for controllable, interactive 3D human motion generation that runs entirely locally. A local Python backend (over gRPC) serves multiple text-to-motion models (T2M-GPT, MoMask, MDM) and writes results straight back into Unity as humanoid AnimationClip assets. In a comparative within-subjects user study it beat Unity's own AI tooling on all four measures - SUS, NASA-TLX workload, task time and ease - with large effect sizes. Graded 77. Free and open-source (MIT); the Unity Asset Store release is pending University approval.
A full-stack habit tracking and goal management web app with social features. Users set numeric goals (e.g. 100,000 push-ups), track daily habits, view AI-powered insights, and compete with friends. Features real-time activity feeds, event planning with Google Maps integration, and a GitHub-style consistency heatmap.
A community-focused web app connecting neighbors who need help with volunteers willing to assist. Users post requests for help (groceries, errands, companionship) and volunteers view them on an interactive map and offer assistance via real-time chat. An evolution of our HackSheffield 9 hackathon project where we won 1st place in the GitHub competition.
An image classification study built by fine-tuning AlexNet and selectively re-training specific convolutional layers, followed by a range of analyses comparing training strategies, layer behaviour, and classification performance.
A set of modelling and analysis studies of natural and dynamical systems, including the Duffing oscillator and a Predator-Prey agent-based model, alongside an exploration of tabular Q-learning.
A real client website delivered end-to-end with a full requirements → prototype → feedback iteration loop, followed by deployment and domain transfer coordination with the client’s previous web provider.
A Next.js based old personal portfolio website, utilising various technologies such as a CMS, Three.js and Shaders.
A Ruby on Rails language learning app made for a real client as a university project. Through requirements and regular meetings we developed an app that met the desired standards.
A complete machine learning pipeline built entirely from scratch in Python for handwritten digit classification, featuring custom SVM and KNN implementations with advanced preprocessing and 99.8% accuracy on noisy test data.
A 3D engine pipeline built from scratch enrirely in python, usiliting rasterzing, loading models from .obj files and using custom vector and matrix types.
A third-place Engineering You’re Hired (EYH) concept project: a decentralised swarm of pipe-crawling robots for inspection and leak detection, designed with failure-tolerant locomotion, robust operational protocols, and clear system-level animations.
This live portfolio you're browsing: an AI-augmented, animation-rich Next.js site with streaming DeepSeek V4 Flash chat, contextual system prompt generation, markdown rendering, and dynamic placeholder suggestion engine.
A mobile based app that runs distilled models on the phone to access AI without internet. It also has personalities, offline maps and survival guides.
A complete Texas Hold'em poker game implementation in pure Haskell featuring multiple AI strategies, comprehensive hand evaluation, and full game mechanics including betting rounds, blinds, and sophisticated tie-breaking systems.