The lab · MIT CSAIL
About the lab
Mantis Discovery is a research group within Mantis, at MIT CSAIL. We use AI throughout our research and publish our results as papers on Zenodo.
25+ Researchers in mathematics, machine learning, physics, chemistry and biology work with us. They include PhD students and olympiad medalists.
In a torus, passing particles go all the way around, while trapped particles bounce back and forth on the outer side. The paper’s model includes both.
Team
Lead
- Founder and lead
Subject directors
- Sushaan KandukooriDirector of mathematics and machine learning
- Shrikrishna RameshDirector of chemistry and biology
Researchers
- Quantum robotics research assistant at Portland State University, using quantum fuzzy logic for balance and decisions in legged robots. Also programs autonomous robots for FIRST Tech Challenge.
- Abhijay Gangarapu
- Single-cell biology. Built single-cell analysis tools into Mantis and integrated the FlyBase platform.
- Designs peptides that cross the blood-brain barrier to target glioblastoma. First author of a paper at IEEE CIBCB 2026.
- Computational biology and genetics. Co-authored a paper in mAbs on predicting antibody purification, and is modeling how shear flow affects Tau aggregation.
- PhD researcher in computational biology. Built ParaSurf and AntiSite to predict protein binding sites, and designs antibodies with generative models.
- Works on knowledge graphs and medical image detection in Mantis. Earlier did light-sheet microscopy research on cortical layers at the MIT and Harvard LCN Lab.
- Optimization theory for machine learning. Proved last-iterate convergence for AMSGrad-AA (under review at TMLR). On the Mantis embeddings team.
- Bioinformatics. Fine-tunes protein language models to predict protein function, and studies how their embeddings capture evolution.
- Devom Brahmbhatt
- Quantum robotics at Portland State University. Found an exact way to measure what an entangling gate adds to a quadruped robot's choices, tested on a real Unitree Go2.
- University student in mathematics and theoretical computer science. Tests whether neural combinatorial solvers beat classical heuristics, on Traveling Salesman instances set on geodesic terrain.
- Machine learning. Recreated Google DeepMind's SynthID, the system that watermarks AI-generated content.
- Human-AI interaction. Showed with a UC Santa Cruz professor that duplicate documents mislead retrieval-augmented AI, and that removing them raises accuracy by 16.6 points.
- Engineer with five years in data science and machine learning, including predictive models for healthcare and behavioral data. In Mantis, works on embeddings, knowledge graphs and agents for healthcare data.
- Computational biology. Built MantisBrain, a gene-expression pipeline on the Mantis API, and worked on population genetics at Stanford's Rosenberg Lab.
- Krish
- Builds knowledge-graph and retrieval pipelines for AI agents in Mantis, and studies working memory in language models with Dartmouth.
- USABO semifinalist. Designs protein binders in a lab at MIT and leads protein and drug discovery work on Mantis.
- Mechanistic interpretability. Showed that rare and non-Western facts settle deeper inside transformers, and built a method to edit them. Emergent Ventures fellow.
- Builds applied AI systems. Compared variational quantum circuits with matched classical models across 105 runs (submitted to arXiv), and co-founded Holmes, an autonomous macOS agent.
- Undergraduate researcher in machine learning: multi-agent reinforcement learning for railway traffic, NLP pipelines, and an EACL 2026 workshop paper.
- Language model safety. Studied at UC Santa Cruz how steering persona vectors changes when models refuse, and how that steering throws off automated attack scorers.
- Taksh Kothari
- UIUC undergraduate applying machine learning to biology and computer systems, with five published papers across cardiology, drug discovery and neuroscience.
Joining the lab
If you’re a student who wants to do serious research with AI, or a researcher who’d like to try this in your own field, write to us and tell us what you work on.