The score so far

I worked with a lean, fast-moving team that punches well above its weight. That meant working across DevOps, infrastructure, data science, and modelling.

Analysing large, messy, time-sensitive datasets comes with its own challenges. I built reliable data pipelines and fast backend libraries alongside the modelling work. I vectorised slow loops with NumPy, compiled the hottest paths with Numba or Cython extensions, and implemented specialised local caching, so analysis that used to keep people waiting came back while the question was still fresh.

I spent the summer helping AI agents find the right context and measuring whether their answers, and the data behind them, could be trusted.

I built end-to-end retrieval workflows for agentic tasks, quality checks for a global data pipeline, and a lighter, cheaper way to evaluate model output that still kept people in the loop. As AI agents have become more widespread, the data they access has become increasingly important. Embedding schemes, search heuristics, and evaluation all help make their answers more reliable.

Underneath that sat backend library work in Rust and Python. These systems need to handle enormous volumes of data. Rust was a clear choice for parallel performance and paired well with Python through PyO3. I also worked with subject matter experts to deploy monitoring dashboards, so the team could watch how all of it behaved in production and spot problems before anyone downstream felt them.

I’m back on the same team, and these days the work leans towards software engineering. I build internal tools, look after our databases, and write the backend developer and AI tooling that the team’s analysts and data scientists depend on.

That includes a Python layer over the bond analytics platform, and a server that lets portfolio managers perform analysis directly from their AI tools.

The aim behind all of it is the same: help the people around me get from a question to a reliable answer as quickly as possible. These tools have helped us ask more questions and better ones.