Selected Work

Examples below describe scope, technical challenge, and delivery role without disclosing confidential information.

Roche / GOSH partnership: multi-workstream applied AI leadership

Problem: establish and deliver multiple AI initiatives in a complex healthcare context with diverse stakeholder groups.

Why difficult: governance, interoperability, clinical workflow constraints, and the need to align technical and non-technical teams.

What I did: led cross-functional workstreams across strategy, prioritisation, technical direction, and delivery execution.

Methods/technology: applied ML/NLP delivery practices, architecture planning, stakeholder operating cadence.

Outcome: sustained progress from opportunity framing into implementable, clinically relevant workstreams.

Problem: unstructured genomics reports limited downstream analysis and clinical usability.

Why difficult: clinical language variation, document heterogeneity, and interoperability requirements.

What I did: designed and delivered an end-to-end NLP pipeline to extract, normalise, and structure genomics information for downstream products.

Methods/technology: NLP/document intelligence, ontology-informed extraction, FHIR mapping, dashboard integration.

Outcome: created a practical pathway from reports to structured outputs that support clinician decision workflows.

Problem: support earlier, better-informed intervention planning for chronic kidney disease progression.

Why difficult: data quality variation, longitudinal complexity, and deployment requirements for clinical usage.

What I did: contributed to model development and shaped a practical path from analysis to deployable clinical support.

Methods/technology: predictive modelling, feature engineering, clinical validation collaboration, deployment planning.

Outcome: advanced from modelling work toward real-world clinical implementation readiness.

Problem: internal teams needed faster, more reliable access to compliance and policy knowledge.

Why difficult: governance expectations, source traceability, and trust requirements for LLM-supported answers.

What I did: led design and implementation of an enterprise-ready retrieval-augmented chatbot aligned to compliance use cases.

Methods/technology: RAG architecture, retrieval pipelines, prompt and response controls, operational handover.

Outcome: delivered an internal knowledge capability with practical guardrails for organisational adoption.

Problem: operational reporting and prescribing insight workflows were manual, slow, and hard to scale.

Why difficult: fragmented datasets, variable definitions, and the need for trusted outputs for decision-making.

What I did: led and contributed to analytics product development, automation design, and stakeholder-focused reporting outputs.

Methods/technology: analytics engineering, dashboarding, workflow automation, product-style iteration with users.

Outcome: improved repeatability and timeliness of insight delivery for service and clinical stakeholders.