Hey, I’m Botond

Cambridge, United Kingdom

I research and develop Machine Learning algorithms. I’m especially interested in Bayesian methods for causal inference and structured deep learning to make models more robust at algorithmic generalisation and reasoning, while also improving their interpretability.

Program synthesis Representation learning Algorithmic reasoning Causal discovery Reinforcement learning

Education

September 2023 — September 2024

MSc Machine Learning

University College London · Distinction

Thesis: Hierarchical Bayesian Program Synthesis for Neural Algorithmic Reasoning · Supervisor: Prof. Mirco Musolesi

October 2020 — June 2023

BSc Physics with Theoretical Physics

Imperial College London · Honours

Thesis: Computational Evolution · Supervisor: Dr. David Clements

Botond at UCL graduation ceremony

About

After completing my Machine Learning MSc at UCL, I am continuing our research on Neural Program Synthesis with Prof. Mirco Musolesi in the Machine Intelligence Lab. I have an undergraduate degree in Theoretical Physics from Imperial College, where I completed my final year project with Dave Clements in the Astrophysics Group. I also spent a summer as a research intern under the supervision of Mark van der Wilk working on Causal Discovery and GPLVMs. I grew up in Budapest, Hungary.

Experience

2025 — Present

ML Research Engineer (Contract)

Sherpa · London

  • Designed and delivered proof-of-concept for LLM integration into a survey analysis product.
  • Built a custom chain-of-thought pipeline that translates natural language queries into JSON requests.
  • Ran comparative experiments to inform feasibility studies and strategic model recommendations.

2024 — Present

Postgraduate Researcher

Machine Intelligence Lab · University College London

Research into neural program synthesis and algorithmic reasoning with Prof. Mirco Musolesi, developing a wake-sleep library learning framework with Transformers.

2023

Research Intern

Imperial College London

Investigated causal discovery via cross-validation compared with Bayesian model selection (GPLVM) under Mark van der Wilk. Funded through Imperial UROP with extensive HPC exposure.

Contact

Email · GitHub · Bluesky · LinkedIn