I'm an AI & Data Science undergraduate at Bournemouth University. I build developer tools, explore learning systems, and run small experiments to understand when an idea actually helps.
My main projects are Compiler, VibeGraph and CogniGraph. Alongside those products, I keep research projects and smaller labs separate below: an experimental result, a learning exercise and a usable tool serve different purposes.
Turns a request into structured prompts and an execution plan. It also includes an advisory tool for checking pull-request readiness, making assumptions, checks and next steps easier to inspect before acting.
Maps Python, JavaScript and TypeScript code into an interactive graph. Source views and AI explanations help connect the visual structure to the actual code, especially when exploring an unfamiliar project.
A browser-based 3D learning tool for exploring brain regions, cognitive pathways and simplified neuron simulations. The current version runs on the client, without a backend or API key. It is an educational exploration tool, not a medical model.
Projects organized around a question, an experiment or a proposed test. Their repositories distinguish measured findings from exploratory ideas and unfinished work.
| Project | Question and current scope |
|---|---|
| SpacingLab | How do repetition timing and selective replay affect knowledge retention during small-model fine-tuning? Includes spacing comparisons, mixed replications and a positive three-pair selective-replay screen, with raw measurements and limitations. |
| Does locality help? | Does a local neural-network learning rule reduce forgetting during sequential learning? Completed experiments examine alternative explanations and measurement limits; they do not establish a generally better learning rule. |
| Entangled AI Learners | An exploratory MNIST notebook studying output-feedback learning between a convolutional network and a multilayer perceptron. A learning experiment, not a general claim about collaborative learning. |
| Rigged Brief | Will an agent notice and report a quietly weakened task requirement? The evaluation setup and pilot work are recorded; the main study is not completed and its hypotheses remain unanswered. |
Smaller developer utilities, simulations and hands-on learning projects. These are not presented as research findings.
| Project | What it explores or does |
|---|---|
| agy-understudy | A live cockpit for following coding-agent runs delegated through Google Antigravity. |
| ctx | Saves a working-directory snapshot and shows what changed when returning to a project. |
| agent-receipt | Reads local agent transcripts to compare requested and actual models, usage and policy checks. |
| SheetCI | Scans Excel workbooks locally for risky formulas and produces audit reports. |
| PaperHunterAgent | A command-line tool for discovering and summarizing quantum-science papers. |
| STDP Neural | An educational spiking-neuron simulation exploring how spike timing changes connection strengths. |
| Autism Screening Classification | A tabular machine-learning workflow exercise using scikit-learn and MLflow. Educational only, not a diagnostic tool. |
- NeurIQ — a personalized-learning project exploring study maps, quizzes and learning feedback.
- AgenticPlace — a local-first workspace for chat, agents, workflows and documents.
I write about AI, learning and experiments on Medium, including the questions behind these projects and what I learned while testing them.