Adaptability

I pick up unfamiliar stacks fast.

This site is proof. I did not know Next.js, TypeScript, or Tailwind a few weeks ago. It is built, deployed, and you are looking at it right now.

Honesty

I am upfront about how I work, including the tools I use.

I used AI as a tool while building this site, the same way I would use a Stack Overflow or documentation. Every architectural decision, every line of content, and every design choice on this site is one I made and can defend.

Pragmatism

I use whatever is the right tool, not the popular one.

On a retail demand forecasting pipeline, a Random Forest model beat complex neural networks like LSTM and GRU by 60% on time series data. I documented that without hesitation.

Curiosity

I do not stop at what works. I want to know why.

When Random Forest outperformed LSTM, a model known for handling sequential data well, I analysed the data itself and found it held no long-term dependency for LSTM to exploit. Model selection should be driven by the nature of the data, not by a model's reputation in a given domain. That reputation is a reason to test a model, it is not a reason to skip testing the others.

First-principles thinking

I try to understand the concept underneath the technique.

A paper encoded time using sine and cosine transformations for a train delay model. Instead of copying it, I worked out why: a clock is not a straight number line. On a number line, 6 is closer to 2 than 11 is. On a clock, 11 is closer to 2 than 6 is, because of its cyclical nature.

Accountability

I take full responsibility for my mistakes and fix them.

On a train delay prediction model, naive outlier removal looked reasonable but increased error by 129%. I caught it, reported it plainly, reassessed the preprocessing strategy, and rebuilt it properly.

Problem solving

I bring solutions, not just effort.

I ranked 1st and 2nd in two machine learning challenges during my master's at the University of Nottingham, and finished runner-up at a C programming bootcamp during my bachelor's.

Maturity

You can trust me with your codebase

Every experiment I run is tracked, not just remembered. On the demand forecasting and bankruptcy projects, every model, parameter, and metric was logged through MLflow before anything was called final, so a change that helps or hurts is provable, not a guess. Deployments run through Docker and CI/CD rather than manual pushes, so what gets tested is exactly what gets shipped. Small changes get committed to GitHub and reviewed step by step, not bundled into one large, unreviewable difference.