MTS - Research (Biosafety Simulations)

We are extending our environments into biosafety, and we need a domain authority to lead that track.

You will turn the everyday judgment of a biosafety professional into tasks, reference solutions, and graders that measure whether a model reasons about biological risk the way a trained expert would. Because biology is dual-use, you will also help define the boundary: where a model should answer fully, where it should hedge, and where it should decline.

• Map real biosafety work into tasks. Risk assessments, biosafety-level assignment and containment decisions, IBC protocol review, exposure and spill response, decontamination and waste streams, shipping of infectious substances, ecotoxicology study design, and environmental release assessments.
• Write expert-level prompts and gold-standard answers. Realistic scenarios with the ambiguity a practitioner actually faces, paired with reference solutions that hold up to peer review.
• Build rubrics and verifiers. Define what a complete, correct, and safe answer looks like, and what an unsafe or insufficient one looks like. Calibrate your rubrics against other experts so grading is consistent.
• Grade model outputs. Score responses for scientific accuracy, regulatory correctness, completeness, and appropriate handling of sensitive information.
• Red-team the boundary. Design adversarial and multi-turn prompts that probe where helpful biology becomes operational uplift. Document failure modes clearly enough that engineers can act on them.
• Shape the environments. Work with our environment engineers to encode workflow state, tools, and reward signals, and review model trajectories for realism.
• Raise the bar for the track. Review tasks written by other contributors and help set the quality standard for biosafety work at Collinear.

• Experience as a biosafety officer, IBC member, or EHS professional. CBSP or RBP certification.
• Ecotoxicology or environmental risk assessment practice, including regulatory submissions.
• Prior work training, evaluating, or red-teaming AI systems.
• Python or basic scripting for working with data and evaluation pipelines.