Selected work

Projects.

Systems that shipped, got adopted, and kept running: fraud detection and forecasting for a global pharma finance function, risk intelligence for a central bank, and GenAI work of my own. Each one below opens into the problem, what I built, and what it changed.

157

automation projects delivered

12,000+

hours saved annually

8

markets in production

Microsoft certified

AI & agents

Retrieval, generation, explainability.

Where the model output has to justify itself before anyone will act on it. The pattern is the same in a consumer prototype and an enterprise decision-support system: retrieve, generate, then show your reasoning.

Automation at scale

Built once, run everywhere.

Finance systems that had to survive eight markets, each with its own currency, calendar, and reporting habits. Standardising the inputs was most of the work; the forecast was the easy part.

Academic & personal builds

Robotics, before any of this paid the bills.

Coursework, not shipped work — kept separate from the case studies above on purpose. The engineering problems were still real: arm kinematics, and getting subsystems designed by seven different people to work as one machine.

157

automation projects delivered at AstraZeneca

The pages above are the ones worth reading end to end. They sit inside a much larger programme: Malaysia's first nationwide digital transformation, where my team upskilled 492 people and took 12,000+ hours of manual work out of the business every year.

The full history →
Contact

Want the detail behind any of these?

Some of this work sits under pharma and central-bank confidentiality, so the write-ups stay at the architecture level. Happy to go deeper in conversation.