Staff ML Research Scientist

Leads end-to-end applied ML research in NLP, LLMs, retrieval, and multimodal models for healthcare AI, driving from experimentation to production deployment with rigorous evaluation and clinician collaboration. Requires 7+ years experience, MS/PhD, and depth in ML areas like PyTorch tooling.

About the job

What You'll Do

Own end-to-end applied research: frame the problem, design experiments, ship to production, and monitor impact against real-world metrics.
Set technical direction across LLMs, retrieval, and multimodal; run ablations/error analysis that change product decisions.
Build evaluation that matters: link offline metrics to online outcomes; define thresholds, monitoring, and rollback.
Partner to deliver with engineering and product—and, when relevant, clinicians/domain experts—to align data, success criteria, and timelines.
Raise the bar by mentoring peers and codifying standards for reliability, safety, and documentation.
Improve the platform (data, training, serving, observability) to speed iteration and ensure reproducibility.
Explore new directions, with computer vision/vision-language work as a nice-to-have for future strategic initiatives.

What We're Looking For

MS or PhD (or equivalent research experience) in Computer Science, Electrical Engineering, Computational Linguistics, Biomedical Informatics, or related quantitative field.
7+ years of applied ML research experience (or PhD + 5 years, or equivalent evidence of Staff-level impact).
Depth in one or more areas: LLMs and NLP, computer vision, speech, recommendation/ranking, retrieval, or multimodal modeling.
Strong experimental rigor: clear hypothesis framing, offline→online linkage, calibration and stratified analyses, ablations that influence decisions.
Proven ability to take models to production.
Hands-on with modern tooling: PyTorch and common experiment/ops tools (for example MLflow, Databricks, Ray, or similar).
System thinking: can choose methods based on constraints, design for observability and rollback, and document decisions clearly.
Collaborative communicator who writes crisp design docs and explains complex ideas to non-specialists; comfortable mentoring peers.

Preferred Qualifications

Health data familiarity, including EHR or imaging.
Experience in one or more areas: clinical NLP or LLMs, computer vision, speech, retrieval or multimodal modeling.
Shipped, measured models in production with monitoring and clear rollback; external or multi-site validation is a plus.
Workflow integration with EHR, RIS, PACS, or reporting systems; PowerScribe or Dragon exposure helpful.
Strong evaluation practices: calibration, slice analysis, and ablations.
Safety and governance in sensitive domains, including PHI handling and HIPAA or FDA-adjacent environments.
Technical mentorship and contributions to team research culture; publications or impactful open-source work.
Practical tooling: PyTorch plus modern ML ops tools such as MLflow, Databricks, Ray, or Triton.