Developer-turned-data-scientist building measurable, data-driven products.

Role

Sole researcher and developer — problem framing, forecasting model, scheduling engine, and UX validation.

Key decisions

• Used Prophet to forecast demand and staffing needs from historical shift data rather than relying on fixed rosters.
• Built a custom constraint-based scheduling engine instead of an off-the-shelf solver, because existing solvers could not hold the real fairness and availability constraints gathered from staff.
• Designed the scheduling UI in Figma and validated the concept with a staff survey before building the working prototype.

Result

Delivered a working system that generates rotas balancing forecasted demand against fairness constraints, validated against real staff feedback rather than assumptions.

Role

Built and deployed independently, alongside day-to-day operational duties.

Key decisions

• Chose WhatsApp as the interface since it required zero training or new habits from staff already using it daily.
• Used Supabase/Postgres as a lightweight, low-maintenance backend suited to a small team rather than standing up dedicated infrastructure.
• Added a Streamlit dashboard for managers to review task history without needing database access.

Result

In active use for logging and tracking shift-handover tasks, reducing dropped handoffs between shifts.

Role

Sole developer — geospatial analysis, backend, and mobile app.

Key decisions

• Used Uber's H3 hexagonal grid system to bucket geography into uniform cells, making coverage gaps and demand density directly comparable across a region.
• Modelled station and grid-cell data in MongoDB to keep geospatial queries flexible during exploratory analysis.

Result

A working mobile app that surfaces candidate siting locations for new charging infrastructure based on grid-cell coverage analysis.

Role

Mobile Application Analyst & Data Analyst — led the rebuild of the patient app and standardisation of EMR data behind it.

Key decisions

• Ran a structured review of complaint data to prioritise fixes by actual patient impact rather than assumption.
• Redesigned core flows in Figma before touching code, validating changes against real complaint themes.
• Standardised over 100,000 EMR records against Miracle EMR to make patient data faster and more reliable to retrieve inside the app.

Result

App Store rating rose from 1★ to 4.6★, user complaints fell by 50%, daily active usage grew 35% within three months, and patient satisfaction scores rose from 60% to 85%.