Applying sparse autoencoders to decode what deep learning models actually learn in genomics. Presented at NeurIPS 2025 Workshop on Multi-modal Foundation Models for Life Sciences.

A Model Context Protocol server that provides access to OpenReview data for research and analysis. Added to the official MCP servers list.

Insertions and deletions comprise 25% of all common variants in the genome, but predicting the effects of indels on expression and epigenome is technically challenging. Clever augmentation techniques fix this.

A fun revamped project on attractors in NNs. Winner of the CoSScience Art Award during my PhD years.