What do 30 years of river monitoring data tell us about England's waterways?

Dug into decades of Environment Agency biomonitoring records: messy, inconsistent, unevenly sampled; to find long-term shifts in macroinvertebrate communities. These tiny creatures are one of the most sensitive indicators of water quality we have.

Built reproducible preprocessing pipelines from scratch and applied Generalised Additive Models to let the patterns speak for themselves.

Can interpretable models help regulators understand what's driving seismic risk?

Industry collaboration: integrating operational and temporal variables from heterogeneous sources to surface what actually matters for risk assessment and communicating that clearly to decision-makers and domain experts.

The interesting challenge here was making the model explainable, not just accurate.

Where is the NHS under the most pressure, and when?

Explored attendance patterns, seasonal demand, and performance against the 4-hour standard across English A&E departments. Built to surface systemic pressures in a way that's accessible — not just to analysts, but to anyone who wants to understand how the health system is coping.

How do you run an experiment that actually means something?

A rigorous experimentation framework: power analysis, experimental design, hypothesis testing, and honest interpretation. The focus is less on the result and more on how you think about evidence: avoiding p-hacking, early peeking, and the quiet dishonesty of underpowered tests.