Projects

Selected projects

Work across machine learning, LLM systems, computer vision, data pipelines and full-stack products, written with notes on the problem, architecture, modelling decisions and implementation trade-offs.

AI Systems · Sensor ML · Industry Research2026

Dogtooth AI Collision Intelligence System

MSc industry dissertation with Dogtooth Technologies. A sensor-driven pipeline that detects harmful infrastructure interactions on autonomous strawberry harvesting robots, reducing downtime without compromising safety.

Python · Sensor ML · Time-series features · Anomaly detection · Severity classification

In progress

Computer Vision · Domain Adaptation · Deep Learning2026

Strawberry Ripeness Detection Under Day-to-Night Domain Shift

Computer vision project tackling performance drop in fruit ripeness classification when lighting shifts from daylight to LED-illuminated polytunnels. Built on the StrawDI_Db1 dataset with a CycleGAN-based domain adaptation pipeline that translates daytime imagery into the night domain for training, evaluated before and after adaptation with accuracy, precision, recall and confusion matrices.

PyTorch · CycleGAN · Domain adaptation · Transfer learning · Model evaluation

Research

LLM Security · Data Loss Prevention · Google London Final2026

PromptGuard AI: Enterprise LLM Data Loss Prevention

Applied AI security platform built during the WMG / Google / NatWest Secure Intelligence Frontier hackathon (advanced to the grand final at Google London). Intercepts prompts sent to public LLMs from the browser and detects sensitive data leakage in real time. Chrome Extension (Manifest V3) for in-browser interception, FastAPI backend on Google Cloud Run for detection and policy enforcement, and a React admin dashboard for analytics and rule configuration.

FastAPI · Cloud Run · Chrome Extension MV3 · LLM classification · React dashboard

Active

Agentic AI · Semantic Search · Full-Stack Product2025

Agentic AI Job Application Copilot

End-to-end automation platform that finds, ranks and helps apply to relevant graduate roles. FastAPI + SQLAlchemy 2 backend with APScheduler-driven Greenhouse scrapers, an embeddings-based semantic matching pipeline that ranks roles against a candidate CV, a Next.js 15 frontend with TanStack Query, and a 28-test suite locking in scraper reliability and matching logic.

FastAPI · Next.js 15 · Embeddings · SQLAlchemy 2 · Pytest suite

Active

Reinforcement Learning · Autonomous Agents · Simulation2026

IBM AI Racing League: Autonomous Driving Agent

Autonomous racing agent in the TORCS simulator built with Python and reinforcement learning. The agent consumes telemetry such as speed, track position and orientation, then produces real-time steering, throttle and braking. Reward shaping and training loops were tuned to improve lap completion and stability; IBM Granite supported algorithm design and debugging.

Python · Reinforcement learning · TORCS · Policy control · IBM Granite

Active

Responsible AI · Model Selection · Client ML2026

Bengazi Radar: Redundancy Risk Prediction System

Client-facing responsible AI project for Bengazi Career Engagement. Trained and evaluated multiple machine learning algorithms for redundancy risk prediction, tuned model hyperparameters, compared performance, and documented the final predictive pipeline with architecture, evaluation and responsible AI considerations.

Python · XGBoost · Model comparison · Hyperparameter tuning · Evaluation

Client project

Data Engineering · Public Health Analytics · Python CLI2025

Public Health Data Insights Dashboard

Python data insights system for malaria epidemiology reports across African countries. Built with a Click CLI, pandas analysis services, SQLite persistence, validation and cleaning utilities, CSV/JSON export, activity logging and a layered domain-driven architecture with a 36-file test suite.

Python · pandas · SQLite · Click CLI · Pytest suite

Active

Sports Analytics · Predictive Modeling · Full Stack2023

EPL Predict: Real-Time Premier League Match Predictions

Full stack web application combining historical match data, live APIs and a statistical / ML predictor for English Premier League matches, with a lightweight AJAX frontend.

Python · Flask · MySQL · REST APIs · Feature engineering

Archived