GenAI & Applied ML · LLM pipelines · MLOps · Predictive modelling
Data Scientist with 4+ years across analytics, machine learning, and Generative AI application development in insurance, logistics, and pharmaceutical domains. I design and deploy LLM-based pipelines, predictive models, and cloud ML infrastructure on Azure using Python, Databricks, MLflow, FastAPI, and GitHub Actions.
I build ML and Generative AI systems that ship — not just notebooks. My work spans LLM-orchestrated pipelines, predictive modelling, and the MLOps stack that keeps them running in production.
Over 5–6 years my career has followed one intentional arc: from SQL, dashboards, and forecasting, through enterprise analytics and automation, into classical ML during my MSc, and now into applied ML and GenAI at production scale. The Bachelor of Pharmacy underneath it gives me a genuine edge in healthcare and life-sciences data — from clinical NLP to pharma demand forecasting.
I care about correctness, deployment quality, and connecting model performance to real business outcomes — grounded in an MSc in Data Science (Distinction) from the University of Essex.
Manual case review of recovery calls was slow and inconsistent. This production pipeline automates it end-to-end — transcription, LLM-based summarisation, sentiment-driven analysis, and automated action generation — freeing agents to act on insights instead of transcripts.
Speech-to-text transcription → LLM API summarisation and intent/sentiment extraction → automated recovery-option and action generation. Served via FastAPI on Azure with MLflow experiment tracking and GitHub Actions CI/CD.
Screening molecules for anticancer activity by hand is expensive and slow. This classification pipeline predicts activity directly from molecular structure, using descriptors and fingerprints to prioritise promising candidates.
Molecular descriptor and fingerprint generation with RDKit → feature selection → classification via scikit-learn ensemble methods, validated with cross-validation to keep results honest on unseen molecules.
Clinical signals for gout are buried in free-text notes. This NLP pipeline classifies gout-related clinical text to surface structured risk signals from unstructured records.
Text preprocessing → TF-IDF vectorisation → supervised classification with scikit-learn, tuned via hyperparameter search and assessed with precision, recall, and F1.
Reactive inventory planning on pharmaceutical e-commerce data led to avoidable stockouts. These forecasting models anticipate demand ahead of time so inventory is planned, not patched.
SQL-based data preparation on transactional and inventory data → time-series forecasting models per product category → results surfaced through Power BI for planning and stockout reduction.
Open to Data Scientist, ML Engineer, MLOps, and GenAI roles where statistical rigor, deployment quality, and business impact all matter — insurance, logistics, healthcare, and enterprise SaaS all of interest.