Founding Research Scientist, Biological Foundation Models

You may be a fit if you

Modeling depth

• Have trained or materially improved large models yourself, in any domain: vision, multimodal, language, speech, protein/molecule, or biology. You know what it takes to get a large training run to converge, what breaks at scale, and how to diagnose it
• Have strong applied ML judgment: representation choices, tokenization or encoding of non-text data, loss design, pretraining vs. fine-tuning vs. post-training trade-offs, and rigorous evaluation
• Are fluent in Python and PyTorch or JAX, and comfortable running experiments on multi-GPU infrastructure
• Have shipped models that people relied on in production or in a live program

Biology fluency

• Have worked hands-on with biological data such as scRNA-seq, Perturb-seq / CRISPR or compound screens, spatial transcriptomics, bulk or cell-free omics, histopathology, or high-content imaging, and understand its noise, batch effects, and failure modes
• Understand the biology behind the data well enough to ask whether a prediction is plausible, to choose sensible baselines and controls, and to talk credibly with pharma scientists
• Have followed or contributed to the virtual-cell / perturbation-modeling literature (single-cell foundation models, perturbation response prediction, in-silico screening) and have opinions about what works and what does not

Working style

• Want to own a project from ambiguous question to partner-facing result
• Are comfortable in a small team where priorities move toward whatever matters most this week
• Communicate clearly with both ML engineers and biologists

We'd be especially excited if you also have

• Trained vision or imaging models at scale and have since applied that to biology (cell painting, histopathology, spatial, microscopy)
• Built or evaluated a virtual-cell or perturbation-prediction model against held-out experiments
• A PhD or equivalent research record in computational biology, ML for biology, or a quantitative field applied to biology
• Worked with pharma or biotech partners and understand what a translational or discovery team needs from a model result
• Experience with cross-species, cell-line-to-patient, or preclinical-to-clinical translation problems
• Experience with LLM post-training (SFT, RL, tool use) applied to scientific reasoning

This role will shape what CellType's models can do. The right person will determine not only how well we model biology, but which problems we choose to solve and how quickly our partners see results they can act on.

This role is based in New York City.

If you have trained models at scale and want to point that skill at biology's hardest prediction problems, we'd love to talk.

A meeting lands on your calendar

If they’re a yes, I book the chat. You show up — that’s the whole job-hunt.