12-day longest streak
--- Claude Code Plugins model-trainer - Autonomous supervised ML training for Claude Code. Hypothesis-driven: parallel experiments in git worktrees, two-stage review pipeline, X-Ray dashboard. Execute-Don't-Eyeball methodology. auto-model-trainer - Long-horizon autonomous…
---
Claude Code Plugins
model-trainer - Autonomous supervised ML training for Claude Code. Hypothesis-driven: parallel experiments in git worktrees, two-stage review pipeline, X-Ray dashboard. Execute-Don't-Eyeball methodology.
auto-model-trainer - Long-horizon autonomous ML training for Claude Code. One command, full autonomy, mechanical convergence. Stop-hook-driven exploration across architecture families, Merkle-chained experiment lineage, subagent-dispatched variant building. Tested on 6 Kaggle competitions, House Prices inside medal zone.
deep-report - Turns a research question into a claim-verified, deep-dive PDF. Multi-agent pipeline with adversarial fact-checking — every load-bearing claim gets a verification sidecar.
runbug - Lets the agent drive frontend dev servers by accessibility role/name (not CSS selectors), replacing Playwright in the inner debug loop with three channels.
optmise - Claude Code plugin for MPS performance optimisation. Bandwidth first, measure before you touch the code. Parallel worktrees for hypotheses, dead-end documentation so the next person doesn't repeat your mistakes.
---
Apple Silicon / MLX Model Optimsation Research
qwen3-mlx - C++23 inference engine for Qwen3-4B on Apple Silicon. 125 tok/s, 1.62× llama-cli throughput, OpenAI-compatible streaming server, no Python runtime. Built by failing five optimisation hypotheses and landing one (quantized tied lm_head bandwidth).
marlin-mlx - C++ MLX inference engine for NemoStation/Marlin-2B video VLM on Apple Silicon. End-to-end video understanding running locally on a MacBook.
decomposer-mlx - Image layer decomposition using Qwen-Image-Layered. MLX 8-bit quantised, 2.3× faster than PyTorch MPS. Breaks images into composable layers for downstream editing.
gemma-streaming - Runs Gemma 3 4B on Apple Silicon inside a 2 GB memory budget by streaming layers from SSD mid-forward. Bit-exact against the dense model (max|Δlogit| = 0), 34 tok/s with a 1B speculative draft. Ships the full five-experiment research log with integrity-locked predictions.
---
Web Apps
CamEvents.org - Aggregates 25 sources of startup, innovation and research events across Cambridge into one filterable page. Zero cost, zero accounts, no monetisation — reached 1000+ users.
CozyGrids.com - Free tool that converts images into printable craft patterns (cross-stitch, knitting, diamond painting, crochet, embroidery) with a full editor and PDF/machine export. CIELAB colour science.
PuzzleDojo - Technical showcase: solves all 7 LinkedIn daily puzzles from a screenshot. Browser-side computer vision (OpenCV WASM), constraint propagation, Algorithm X, and BGE embeddings.
---
All projects are personal, maintained in my own time. Not affiliated with or endorsed by my employer.
---
🏢 Previously
AstraZeneca - four-year rotational programme spanning technology innovation, cyber security, learning and development, AI and enterprise transformation, commercial IT and global commercial. Helped shape enterprise programmes including the award-winning Generative AI accreditation for 10,000 colleagues, winning Gold at The Learning Awards 2025. Advised commercial leaders on AI opportunities. Designed and delivered global hackathons engaging a combined 490+ participants. Scaled the TH!NK neurodiversity ERG from ~20 members to 400+ across 9 countries.
---
GitHub Stats
---
-
personal ★ PINNED
A radically simplified, experimental portfolio project. This version is a clean slate: minimal, fast, and focused on custom code: no templates, no bloat, just a direct showcase of my front-end skills and creative coding.
CSS ★ 0 1mo agoExplain → -
model-trainer ★ PINNED
Skill-driven autonomous ML model training for Claude Code. Hypothesise, design experiments, train, review, and report with Execute-Don't-Eyeball methodology.
JavaScript ★ 2 2mo agoExplain → -
qwen3-mlx ★ PINNED
C++23 inference engine for Qwen3-4B on Apple Silicon. 125.77 tok/s (1.62x llama-cli). OpenAI-compatible streaming server. No Python, no llama.cpp, no Ollama.
C++ ★ 0 1mo agoExplain → -
runbug ★ PINNED
Claude Code plugin: thin runtime channel for driving frontend dev servers by accessibility role/name. Composes with the superpowers plugin.
Shell ★ 0 2mo agoExplain → -
-market-trading-dashboard
A real-time forex market monitoring dashboard built with Python, providing live currency pair analysis, technical indicators, and visual market insights.
Python ★ 1 1y agoExplain → -
Hook12aaa
GitHub profile README
★ 0 3d agoExplain → -
gemma-streaming
Run Gemma 3 4B on Apple Silicon with a bounded decode footprint, streaming layers from SSD
Python ★ 0 3d agoExplain → -
kaggle-introverts-ps-s5e7
Introvert Classification - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-stellar-ps-s6e6
Stellar Classification - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-flood-ps-s4e5
Flood Prediction - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-scores-ps-s6e1
Student Exam Score Prediction - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-steel-ps-s4e3
Steel Plate Defect Prediction - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-irrigation-ps-s6e4
Irrigation Need Prediction - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-mushrooms-ps-s4e8
Binary Prediction of Poisonous Mushrooms - auto-model-trainer solution
★ 0 3d agoExplain → -
kaggle-health-ps-s6e7
Student Health Risk prediction (Playground Series S6E7) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB balanced_accuracy 0.94967.
Python ★ 0 3d agoExplain → -
kaggle-calories-ps-s5e5
Calorie Expenditure prediction (Playground Series S5E5) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB RMSLE 0.05965.
Python ★ 0 3d agoExplain → -
kaggle-heart-ps-s6e2
Heart Disease prediction (Playground Series S6E2) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB ROC-AUC 0.95332.
Python ★ 0 3d agoExplain → -
kaggle-tps-jan-2021
Tabular Playground Series Jan 2021 regression -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB RMSE 0.70308.
Python ★ 0 4d agoExplain → -
kaggle-bank-churn-ps-s4e1
Bank churn prediction (PS S4E1) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB ROC-AUC 0.89102.
Python ★ 0 4d agoExplain → -
kaggle-cirrhosis-ps-s3e26
Cirrhosis patient survival (PS S3E26) -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB log_loss 0.41130.
Python ★ 0 4d agoExplain → -
kaggle-house-prices
House Prices regression -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB RMSE 0.12556.
Python ★ 0 4d agoExplain → -
kaggle-spaceship-titanic
Spaceship Titanic transported prediction -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB 0.79798.
Python ★ 0 4d agoExplain → -
kaggle-titanic
Titanic survival prediction -- created by auto-model-trainer, a Claude Code plugin for autonomous ML training. Public LB 0.76076.
Python ★ 0 4d agoExplain → -
auto-model-trainer
Long-horizon autonomous ML training plugin for Claude Code. 12 skills, two-tier mechanical convergence, Merkle-chained experiment lineage. Tested on 6 Kaggle competitions.
Python ★ 0 7d agoExplain → -
puzzle-solvers
Solver, generator, and web app for 7 LinkedIn puzzle games. Constraint propagation, exact cover, Tarjan's bridges, backbite algorithm, OpenCV vision pipeline
Python ★ 0 28d agoExplain → -
decomposer-mlx
Image layer decomposition using Qwen-Image-Layered on Apple Silicon. MLX 8-bit, 2.3x vs PyTorch MPS.
Python ★ 0 1mo agoExplain → -
marlin-mlx
C++ MLX inference engine for NemoStation/Marlin-2B video VLM on Apple Silicon
C++ ★ 0 1mo agoExplain → -
optmise
No description.
Shell ★ 0 1mo agoExplain → -
deep-report
Claude Code plugin that turns a research question into a claim-verified, deep-dive PDF.
JavaScript ★ 0 1mo agoExplain → -
craft-gate
Per-comment gate for Claude Code: KEEP, STRIP, or REWRITE every comment Claude writes against a craftsmanship discipline
Shell ★ 0 2mo agoExplain → -
Kaggle_NFL
No description.
Python ★ 0 4mo agoExplain → -
HF_kaggle
This is my first time using the XGBoost + CatBoost blend approach, but I wanted to give it a go and see if we can score higher on this exercise to find what features work best. It was difficult but fun!
Python ★ 0 4mo agoExplain → -
titanic_LightGBM_Bayesian
No description.
Python ★ 0 4mo agoExplain → -
personal_webpage
My personal website about me. Just some random elements to descripe who i am built in Next.js with d3
JavaScript ★ 0 1y agoExplain → -
json_parser
A lightweight, high-performance JSON repair utility designed specifically for handling malformed JSON, with special consideration for LLM output processing.
Python ★ 0 1y agoExplain → -
console_outlook
A Python-based solution for monitoring multiple Outlook folders simultaneously on macOS, with real-time notifications for new emails.
Python ★ 0 1y agoExplain → -
artist_web_builder
based on the kevin beacon for the music industry
★ 0 1y agoExplain → -
Orangesauce
No description.
★ 0 9y agoExplain → -
food_truck
food delivery vehicle using python
Makefile ★ 0 4y agoExplain → -
AstraZeneca-opencv-training
This is for the outreach for students AstraZeneca will be doing in the coming months. The project is public facing.
Python ★ 0 5y agoExplain → -
Scammer-Spammer ⑂
A webform spammer targeted towards wasting scammer's time.
★ 0 5y agoExplain → -
Small_Project_Showcase
This is all of my small projects
Python ★ 0 5y agoExplain → -
text_based_adventure_game
No description.
Python ★ 0 5y agoExplain → -
render
So...I used opencv
Python ★ 0 5y agoExplain → -
UseFull_Tools_Python
This is my finished scripts which can help any beginner work with python. I built this orginally to help anyone in my class that needed to know different methods. I also tried to improve public methods like my main menu script. Enjoy!
Python ★ 0 6y agoExplain →
No repos match these filters.