Principal AI Scientist (Polytope Bio)

Astera Institute•San Francisco, NY

6d•Hybrid

About The Position

Polytope Bio is a residency project at Astera that is building the missing piece: a post-training engine that closes the loop between frontier AI models and high-throughput biology. Our work will power new applications in generative biology by aligning frontier AI models directly to experimental measurements of what actually folds, binds, and functions. We are looking for a Principal AI Scientist to help launch our AI research program. You will be joining at the point of maximum leverage: early enough to shape the scientific direction, the modeling approaches, and the training strategy. The project is resourced with significant compute, financial runway, and the ability to generate large-scale prospective biological datasets. The researcher in this role will work hands-on to develop and publish new reinforcement learning methods and bio AI models leveraging datasets created by our unique high-throughput biology platform. Longer-term, we are seeking a candidate who is driven by the prospect of growing their leadership of the AI research program. While our immediate focus is on collaborative research strategy and execution, the candidate will have the potential to develop into a co-founding leadership position in the event of a successful spinout.

Requirements

PhD in machine learning, computational biology, or a related field with a minimum of 2 to 5 years of industry research experience post-PhD (accomplished researchers without a PhD are also encouraged to apply).
You have trained models from scratch, not just fine-tuned or called APIs. You have owned real training runs, know where they break, and know how to debug them.
Hands-on experience with generative diffusion models and/or transformer architectures.
Deep familiarity with modern reinforcement learning and preference-optimization methods for deep learning.
A track record of strong research via publications, open-source work, shipped models, or equivalent evidence that you drive results.
You are highly self-directed but thrive in a tight-knit, collaborative founding partnership.
Comfort operating with ambiguity and research ownership. You are excited to build the infrastructure and grow the team.

Nice To Haves

Familiarity with biological research (protein modeling, sequence models, structural biology, or adjacent areas).
Experience building and scaling training infrastructure on large GPU clusters.
Past team management and technical leadership experience.

Responsibilities

Co-direct the research program: Partner directly with the founder to shape the scientific vision, research direction, and technical execution.
Develop RL feedback loop: Design, implement, and improve model post-training methods that translate high-throughput biological measurements into direct reward signals for biological language models.
Hands-on engineering : You will work directly with the technical founder to architect model training infrastructure and build, run, and debug models, training loops, and evaluation metrics.
Bridge wet/dry lab: Partner with the experimental team to ensure that what we measure in the lab and what the models learn are designed as a single, cohesive system.