Foundation Model Scientist – Biological Reasoning & Training

Output Biosciences; New York (NY); On-site; USD 180,000 - 260,000; Full time

Benefits offered by this job

Competitive salary and equity

Excellent medical, dental, and vision

Job summary

Output Biosciences is building a biological reasoning foundation model. You will advance core architecture and training for a system that learns biological reasoning from diverse data modalities. The role spans research to training, designing architectures and objectives, and evaluating learned representations.

You will own pretraining end-to-end, including experiment design, distributed multi-GPU training, and scaling. The team values rigor, reproducibility, and impact on biology.

Qualifications

• PhD with 2+ years post-doctoral or industry research in representation learning and model pretraining.
• Alternatively, a Bachelor's or Master's with 5+ years hands-on research in representation learning and model pretraining.
• Strong publication record at top venues (NeurIPS, ICML, ICLR) with contributions to pretraining methods, self-supervised learning, or foundation models.
• Hands-on experience pretraining large models on diverse, heterogeneous data, including designing training objectives and scaling training infrastructure.
• Proficient in Python and PyTorch, and experience training models on distributed multi-GPU infrastructure.
• Ability to own the full research-to-training pipeline: design methods, train and ship models.
• Produce production-quality code that is well-tested and maintainable, with version control and code review.

Responsibilities

• Push forward the architecture and training objectives of our foundation model, designed for biological reasoning.
• Develop methods for the model to learn across multiple biological data modalities simultaneously, building unified representations of molecular biology.
• Extend the model's reasoning capabilities across biological phenomena, pushing what it can predict and understand about binding, molecular properties, and biological function.
• Own pretraining end-to-end: experiment design, distributed training on multi-GPU clusters, hyperparameter optimization, and iteration.
• Design evaluation frameworks that measure whether the model has learned real biological reasoning, not just statistical patterns in training data.

Skills

Python

PyTorch

Distributed training

Experiment design

Code quality

Education

PhD

Masters or Bachelor's

Tools

Git

Jupyter

Job description

Output Biosciences is building a biological reasoning foundation model. You will advance core architecture and training for a system that learns biological reasoning from diverse data modalities. The role spans research to training, designing architectures and objectives, and evaluating learned representations.

You will own pretraining end-to-end, including experiment design, distributed multi-GPU training, and scaling. The team values rigor, reproducibility, and impact on biology.