Kaggle · Live demo

Disaster Tweets Classification

During a crisis, disaster-related tweets drown in noise. Responders need them filtered automatically - but short, unstructured text makes that a hard classification problem.

  • DistilBERT
  • Transfer learning
  • Transformers
  • Python

83%+ Test accuracy · live demo on Hugging Face

NLP research · NHS (UK)

Patient Opinion Mining from Feedback Comments

NHS trusts collect thousands of free-text patient comments but had no scalable way to tell which aspects of care - staff attitude, waiting time, environment - patients were praising or complaining about.

  • Machine learning
  • NLP
  • n-gram features
  • Topic classification
  • R

NHS Study Corpus & methods published in Health and Social Care Delivery Research

Production AI · Robotic Marketer

AI-Powered Marketing Platform

A direct marketing platform needed to generate on-brand, customised content at volume - keyword and manual workflows couldn't keep up.

  • GPT fine-tuning
  • Claude / OpenRouter / DeepSeek
  • OpenAI embeddings
  • Flask
  • Redis Queue
  • AWS Lambda
  • Django

ML Architecture ML marketing platform architected end-to-end

Healthcare ML · HealthcareLogic

Hospital Length-of-Stay Prediction

Hospitals plan beds, staffing and resources around how long patients will stay - but the existing length-of-stay model wasn't accurate enough to base those decisions on.

  • Python
  • scikit-learn
  • Keras
  • EHR data
  • Feature engineering

+47% Improvement over the existing LoS model