Member of Technical Staff, Benchmarking

GenMD; United States; Remote; USD 150,000 - 210,000; Full time

Job summary

GenMD, a Stanford spin-off, builds AI and data infrastructure for healthcare. You’ll design clinical benchmarks for reasoning and longitudinal histories, and build evaluation pipelines that run at scale to compare models and training runs on real-world data.

Collaborate with clinical experts and partner AI teams to turn benchmark results into research papers and products. We value practical judgment, strong Python/data skills, and a bias for rigorous evaluation.

Qualifications

• Hands-on experience evaluating or training LLMs (PyTorch, Hugging Face or similar).
• Strong Python and data engineering skills; comfortable with large tabular and text datasets (SQL/BigQuery a plus).
• Research taste: you can design an evaluation that measures what matters and resists shortcuts.
• B.Sc. or higher in CS, ML, NLP, biomedical informatics or a related field; LLM/NLP publications preferred.
• About 4 hours of daily overlap with Pacific Time (roughly 9 AM to 1 PM PT).

Responsibilities

• Design and maintain clinical benchmarks for reasoning, longitudinal patient histories, multimodal records and tool use.
• Build evaluation pipelines end to end: runs, scoring, dashboards and reports that compare models and training runs at scale.
• Create rubrics, automated graders and LLM-as-a-judge systems, and calibrate them against clinician review.
• Run systematic failure analysis: categorize failure modes, measure how often they occur and feed the results into dataset design and post-training.
• Work with clinical experts and partner AI teams to turn benchmark results into research papers and products.

Skills

LLM evaluation/training

Python & data engineering

Research-minded evaluation

Timezone overlap

Education

B.Sc. or higher in CS/ML/NLP/biomedical informatics

Tools

SQL/BigQuery

Job description

About GenMD

GenMD is a Stanford spin-off building AI and data infrastructure for healthcare. We grew out of years of work inside Stanford Medicine, alongside world-class researchers and clinicians, and we now work with health systems and data partners on large longitudinal datasets: claims, health records, imaging and clinical notes. We make that data usable for researchers and frontier AI, while patient privacy stays protected.

We're just coming out of stealth, already revenue-generating and well funded. We're intentionally small, move fast, and are focused on building something fundamental. Our CEO is a former Associate Director of AI at Stanford Medicine. See how we work: https://genmd.ai/careers

About the role

Frontier models still fail on real clinical work, and most of those failures never show up in public benchmarks. You'll build the benchmarks and evaluation systems that show where models break on real-world healthcare data, then turn those findings into the data that closes the gap. We call the loop measure, build, re-test.

What you'll do

• Design and maintain clinical benchmarks for reasoning, longitudinal patient histories, multimodal records and tool use.

• Build evaluation pipelines end to end: runs, scoring, dashboards and reports that compare models and training runs at scale.

• Create rubrics, automated graders and LLM-as-a-judge systems, and calibrate them against clinician review.

• Run systematic failure analysis: categorize failure modes, measure how often they occur and feed the results into dataset design and post-training.

• Work with clinical experts and partner AI teams to turn benchmark results into research papers and products.

What we're looking for

• Hands-on experience evaluating or training LLMs (PyTorch, Hugging Face or similar).

• Strong Python and data engineering skills; comfortable with large tabular and text datasets (SQL/BigQuery a plus).

• Research taste: you can design an evaluation that measures what matters and resists shortcuts.

• B.Sc. or higher in CS, ML, NLP, biomedical informatics or a related field; LLM/NLP publications preferred.

• About 4 hours of daily overlap with Pacific Time (roughly 9 AM to 1 PM PT).

Nice to have

• Experience with clinical data (EHR, claims, FHIR/OMOP, clinical notes) or medical benchmarks.

• Experience with RL from verifiable rewards, reward modeling or post-training data.

Compensation

$150,000 – $210,000 per year, plus equity. Final offer depends on experience and location.