About

Intro

Hello, I am a PhD student with a background in Physics and Chemistry. I am currently studying at the University of Oxford, on the AIMS program.

I am very interested in applications of ML to science, particularly to the problem of climate change. Specifically, the intersection of ML and physical modelling with PDEs, particularly in the context of GCMs, ESMs and IAMs. Right now I am particularly interested in: applications of deep learning to PDEs (PINNs, NeuralODEs…), Reinforcement Learning for dynamical systems and Bayesian Deep Learning. If you are a researcher in these areas, I would love to hear from you!

I previously was working as a Machine Learning Engineer at Carbon Re. a startup that uses machine learning to help heavy industry companies reduce their carbon footprint.

Education

PhD Autonomous Intelligence and Machine Systems, Oxford University, 2024-2028

Ongoing, research interests relate to intersection of ML, Science specifically in the context of modelling real world systems.

MSc Machine Learning, University College London, 2021-2022

Distinction, 84%
Studied a wide variety of topics including NLP, RL, Bayesian Deep Learning, Variational inference. My dissertation was about using RL in Integrated Assessment Models, work that I then published at NeurIPS 2023.

BSc Astrophysics and Physical Chemistry, University College London, 2018-2021

First Class Honours, 79%
Starting from the BSc Natural Sciences, I specialised over time into Astrophysics and Physical Chemistry. Got to learn MatLab and some really interesting stuff about the large and small scales of our universe. Did you know that there is 6 times more dark matter than regular matter, and yet we've never directly detected it?

Experience

Machine Learning Engineer, Carbon Re, 2022-2024

I was tasked with making ML models based on high dimensional industrial time series. Most of the focus is on cement data which involves complex physical interactions.
The work also involved maintaining a complex production environment. The pipeline consists of many steps: requesting live data from an API or MQTT (depending on the client), preprocessing the data with AWS lambdas, storing it in AWS timestream. This data is then fed to the inference service which loads ML models from WandB and sends the results to the online web-platform for the clients to view.

Data Science Intern, SuccessData, 6/2021-9/2021

• Development of an NLP-based tool for ENGIE during a 3-month internship.
• Used the Reuters API as well as FinBERT, spaCy, flair in a live pipeline, the results were uploaded on our platform for the Credit Risk team at ENGIE to view.
• Fine-tuning was done on labelled articles from the ENGIE’s Credit Risk analysis team.