We work on AI for Science โ mostly molecular modeling, multimodal foundation models, and the benchmarks / datasets / evaluation tools that go with them.
Breaking Bad Molecules: Are MLLMs Ready for Structure-Level Molecular Detoxification?
A benchmark for molecular toxicity repair with MLLMs โ 660 toxic molecules, 11 repair tasks, multimodal inputs (SMILES + 2D structures), plus the companion ToxiEval evaluation framework.
๐ Dataset | ๐ Paper | ๐ฌ Code
ToxiEval-ZKP: A Structure-Private Verification Framework for Molecular Toxicity Repair Tasks
A structure-private, verifiable evaluation framework for molecular toxicity repair built on zero-knowledge proofs.
๐ Paper | ๐ฌ Code (coming soon)