Highly Experienced Structural Biologists

About the role:

We're partnering with a leading AI lab to evaluate how a frontier model performs on long-horizon protein-design problems. We're running a small, hands-on project with senior protein-design experts to author and evaluate well-scoped design problems — and to iterate live with our team to shape what good looks like.

What you'll do:

1. Brainstorm 2–3 candidate problem ideas (rough descriptions, ~1 page each). Examples of the shape we're looking for:

2. Iterate with us until we align on a task structure

3. Write up the chosen problem: target/inputs, allowed tools and information sources, what success looks like, and — most importantly — how you'd evaluate the model's design decisions and reasoning along the way.

4. Review the model's trajectories (likely 5–20 per problem) and grade design choices + reasoning quality at each step.

Critical fit criteria (please address these in your application)

• You're a working protein-design expert (binder design, FEP/ddG, enzyme engineering, structure-guided med chem, etc.) — PhD or equivalent industry track record.
  • You can showcase this expertise from many different angles. Materials Science, Computer Science, Biophysics, Force Field Chemists. If you work downstream or upstream from protein structures and have the expertise, please apply!
• You can pose a long-horizon design problem whose optimal answer you don't know but where you're still confident judging whether a given trajectory is sound.
• You're comfortable evaluating reasoning quality, not just final outputs. Including cases where the model has full internet/tool access or a constrained toolset.
• You can iterate quickly (multi-day turnaround on drafts) for ~2 weeks.

Time commitment:

• ~15–30 hours over 2–4 weeks (problem authoring + iteration + trajectory grading). There is potential for expansion.

Compensation

• $150–220/hr. We want experts who’ve shipped molecules, maintained or contributed to widely-used tooling, and/or made an impact on the field.