Research intern

We build novel sparsification and tokenization methods that unlock microbiome data at a scale nobody has reached before, train foundation models for microbial communities, and carry those discoveries through to frontier research and clinical outcomes. The mission is enormous. We take it on because we believe a small group of ambitious people with real ownership can do things that look impossible from the outside. About the team Mission-driven researchers and engineers out of MIT, Cambridge and Harvard Medical School, and from companies including J&J and IBM Research. We build frontier methods that hold up at scale. Every person here is a scientist. Every person here writes code. We publish at the leading AI × biology × microbiome venues and stay active across the major AI conferences. Requirements You believe you can learn almost anything in two weeks — and you prove it now and then without being insufferable about it. You know your way around modern training pipelines for large models: RL, SFT, evaluation. You’re comfortable with torchrun, accelerate and multi-node training — or certain you’ll get comfortable quickly. You’re resourceful about data, whether that means cleaning it, generating it, or reframing the problem so you need less of it. You prefer simple answers to hard problems, but you’ll go into PyTorch internals or CUDA when simple stops working. You express ideas clearly. Docs, technical reports and internal memos are part of the work here, not an afterthought. You can turn an insight into a clean user flow and polished UI that customers actually enjoy using. You’ve read most of our papers (see Publications) — in particular: A. E. Gollwitzer, D. A. Subramanian, I. Tucker, and G. Traverso, “Steering the evolutionary game: Hierarchical control of therapeutic resistance in cancer treatment,” in Proc. NeurIPS 2025. A. E. Gollwitzer, D. A. Subramanian, I. Tucker, and G. Traverso, “MetaOmics-10T: The foundational dataset to unlock causal modeling of microbial ecosystems,” in Proc. NeurIPS 2025 AI for Science Workshop, 2025. What we look for People who never quite feel finished. Better always looks reachable. A tolerance for how intense — and sometimes demoralizing — startup work gets. You can spend weeks on an idea and find it moved nothing. The only real failure is learning too slowly. Fast iteration: prototyping, testing and problem-solving alongside Engineering, Product and small cross-functional teams. A talent for creating clarity — communicating design work well and building genuine empathy for our customers across the company. Comfort with scrappy teams. Ship the skateboard, not the car. Your work reaches real patients, not leaderboard positions. Frontier methods only. Everything we build has to scale in the real world — no toy benchmarks. Direct collaboration with world-leading researchers and advisors in the field. Our interview process is fast. After the initial phone screen, we schedule a 45-minute follow-up within 48 hours. If that goes well, we make an offer contingent on a take-home technical and references. That’s it. #J-18808-Ljbffr Norobots Co