Selected Work

Healthcare · Hospital · Regulated AI

AI-driven clinical data integration and visualisation for specialty-specific insights

Challenge: A major hospital needed to unlock clinical insight trapped in unstructured reports and disconnected systems, and bring it into clinicians' day-to-day workflows.

What I led: A strategic AI initiative aligning stakeholders from C-level executives to data scientists and engineers. Brought together NLP pipelines for extracting structured data from PDF reports, FHIR-based interoperability, and clinician-facing dashboards into one specialty-specific solution that addressed real clinical needs.

Outcome: A deployed capability addressing real clinical needs and contributing to published work.

Watch the talk

Healthcare · NLP · Interoperability

Genomics NLP pipeline to FHIR and clinician-facing tooling

Challenge: Genomic variant information was locked inside unstructured PDF reports, slowing clinical use and requiring heavy manual review.

What I built: An end-to-end pipeline extracting genomic variants from unstructured PDFs, mapping outputs to FHIR, and feeding a clinician-facing dashboard. This work formed part of a broader hospital data transformation initiative.

Outcome: Significantly reduced manual review effort and created a reliable route from report intake to clinical decision support.

HL7 FHIR

Enterprise AI · LLMs · Governance

Enterprise RAG / LLM delivery in a regulated setting

Challenge: Internal teams needed trustworthy answers to compliance and policy questions within their existing workflows, with strict governance and traceability expectations.

What I led: Architected and shipped an enterprise compliance-focused assistant with strong governance, traceability, and source attribution requirements. Led product and delivery direction through to rollout to approximately 1,500 internal users.

Outcome: Hands-on delivery of a modern AI system in a real organisational setting where control, reliability, and adoption mattered as much as model capability.

LangChain RAG overview

Telecommunications · Commercial ML

Telecom uplift modelling for call reduction

Challenge: A telecommunications client needed to reduce inbound call-centre demand driven by bill shock, where customers receive a higher-than-expected bill.

What I built: An uplift modelling solution combining propensity modelling with treatment-effect estimation to identify customers most likely to call and recommend the most effective proactive interventions, such as billing alerts and personalised messages.

Outcome: Validated through A/B testing and used to support more targeted, timely customer engagement.

Accenture case study

Financial Services · Predictive ML

Delinquent invoice prediction

Challenge: A financial-services client needed a more reliable way to flag invoices at risk of non-payment so collections effort could be prioritised.

What I built: Applied machine learning to the existing prediction approach, improving the model and integrating results into a client-facing decision workflow.

Outcome: Improved delinquent invoice prediction performance by 10% over the earlier model, helping support more reliable identification and prioritisation of at-risk invoices.

Property Analytics · Deployed ML Product

Cyprus Automated Valuation Model (AVM)

Challenge: The Cypriot property market is characterised by low transaction volume and variable data quality, making automated valuation unusually hard.

What I built: An ensemble-based automated valuation model tailored to the local market, developed in collaboration with an experienced property valuator and incorporating external data sources including satellite imagery.

Outcome: A highly accurate production valuation model deployed on Google Cloud Platform.

Insurance · NLP

Automated clause detection in policy documents

Challenge: Reviewing insurance policy documents for specific clauses was slow, repetitive, and prone to inconsistency.

What I led: A team of two data scientists developing NLP models for clause detection using Hugging Face models and modern NLP practices, demonstrating how AI could automate parts of the document review process.

Outcome: A proof of concept that clearly demonstrated the potential for improving efficiency and consistency in policy review.