About Me
Hi! I’m Faraaz, an AI and Health Researcher at Terra API, where I aim to find interesting insights from health-wearables data using statistical and machine learning tools. I was previously at University College London, where I worked under the supervision Professor Edina Rosta. My work here focused on Reinforcement Learning(RL) for enhanced sampling in Molecular Dynamics(MD) simulations.
Much of my current interests revolve around using ML techniques to accelerate scientific discovery, particulary in health and drug discovery. I am also broadly interested in how ML can help fields like Physics Education and Quantitative Finance.
Current Projects
Reinforcement Learning for Molecular Dynamics Simulations – The nature of MD simulations often makes them slow and prone to getting stuck at unimportant states. Along with the Rosta Group, I’m developing a method that uses RL to alter free energy surfaces, hence ehancing sampling of metastable states we would like to study. We’ve managed to get this to work on analytical 1D surfaces and are currently working on real systems like Magnesium-Phosphate unbinding.
Molecular Embeddings Consistent with Tanimoto Similarity – Molecular embeddings are an important area of work in molecular machine learning. I’m currently working on a project that uses contrastive learning to make these embeddings consistent with Tanimoto Similarity. Current testing on the BACE benchmark outperforms SOTA constrastive models such as MolCLR.
Noise2Noise for Photoplethysmography Data – Most health wearables use PPG signals to measure blood volume changes and present health information. However, this data can be quite noisy. Modern ML methods require clean labels to learn how to denoise signals—data that is not easy to acquire. I’ve managed to show that through synthetic noise and self-supervised learning we can denoise these signals without clean labels! (Preprint will be out soon.)
Publications
Using large language models for grading in education: an applied test for physics – Physics Education, Volume 60, Number 3
Variation in Sleep Duration Across Latitudes and Countries: A Bayesian Hierarchical Analysis of Wearable Data – Preprint
Articles
Understanding Water: A Deep Dive Using Machine Learning
Having struggled to create an accurate functional to describe molecules such as water for years, physicists have turned their attention to how tools like ML can help with the task. In the process, DM21 is born.
How to Train Your Fusion Reactor: Letting Tokamaks Learn Stability
Fusion Reactors, while impressively useful for renewable energy production, often run into chaos due to plasma disruption events. A group of scientists tackle this problem using deep reinforcement learning.
Education
MSci, Physics
University College London | Sep 2021 – Jul 2025
Hobbies
I like to write! Feel free to check out my writing account on Instagram. Additionally, I like to go to the gym, play football/soccer, and boulder from time to time.
Faraaz's projects 