Consulting work at Slalom (2020–present) across financial services, energy, healthcare, and telecommunications. Projects covered AI systems, text analytics pipelines, and data platforms.
Built question-answering systems over large document collections for a major financial institution and a leading equipment rental firm. Each deployment adapted to the sector's data types, compliance requirements, and end users.
Three engagements applying text classification, sentiment analysis, and topic modelling to different operational problems. Work included product taxonomy harmonisation for a global consumer goods company, safety theme extraction from incident reports for an energy utility, and document classification for legal e-discovery.
Telecommunications: network predictive maintenance. Built a device health scoring system to proactively identify at-risk network infrastructure before service disruption, replacing reactive engineer dispatch with planned preventive maintenance. Validated in one region and extrapolated nationwide. Projected savings: $8M+ annually, ~30,000 fewer support calls, ~6,000 fewer service visits.
Healthcare: data lake implementation. Six regional health information exchange networks aggregated patient data from providers across the state: hospitals, clinics, screening centres, and community health organisations. The pipeline was built and maintained by a team of four data engineers and a lead architect. It ingested 2 million records per day covering clinical records, screenings, and patient assessments, all in FHIR (Fast Healthcare Interoperability Resources) format, the standard for exchanging electronic health records across care systems.
Snowflake task orchestrations and parameterised stored procedures managed batch ingestion and export. Data quality checks and transformation rules ran across all source systems before records entered the lake. Compliance controls for sensitive patient data were applied throughout. A legacy JavaScript codebase was refactored into modular Python and SQL, making the pipeline easier to maintain and extend.
Outputs ran on daily, weekly, and monthly schedules depending on data type and recipient, supporting state health programme analytics, business analytics for health stakeholders, and follow-up workflows for screening centres. Programme analytics covered ~5 million residents across the state.