harsh agrawal 🧬

I'm a Master's student in Bioengineering at Imperial College London, interested in applying machine learning to biological problems. For my thesis, I'm training a multimer-native Protein Language Model at Softnanolab.

harshagrawal.1312 [at] gmail [dot] com

publications

An Energy Landscape Approach to Miniaturizing Enzymes using Protein Language Model Embeddings
Jakub Lála, Harsh Agrawal, Fanfei Dong, Jude Wells, Dr. Stefano Angioletti-Uberti
bioRxiv preprint, March 2026
paper doi

experience

projects

JESPR architecture diagram
JESPR: Joint Embedding Pre-training for Protein Structure-Sequence Representations
Aditya Tandon, Harsh Agrawal
Applied a JEPA-style contrastive learning approach to learn joint-embeddings of protein sequences and structures.
DPN HRV ECG signal
Diabetic Peripheral Neuropathy Detection using HRV
Adam Lounis, Harsh Agrawal, Henry Hollingworth, Dr. Martyn Boutelle, Mir Wajih Tawshif, Povilas Sauciuvienas, Dr. Richard Kitney, Rodi Rahman, Sheshpriya Gadiya
Benchmarked nine R-peak detectors on 24-hour ECGs and tested whether HRV alone could separate diabetic peripheral neuropathy from diabetes-only patients — classifiers tied, but three HRV metrics shifted significantly, hinting at early autonomic involvement. Supervised by Prof. Richard Kitney, OBE.

talks & achievements

miscellaneous

Ranking all shows I've watched.

On the beauty of the known and the new (Substack).