Lead Research Scientist, Foundation Models and Agentic Systems - Optum AI

Optum is a global organization that delivers care, aided by technology to help millions of people live healthier lives. The work you do with our team will directly improve health outcomes by connecting people with the care, pharmacy benefits, data and resources they need to feel their best. Here, you will find a culture guided by inclusion, talented peers, comprehensive benefits and career development opportunities.

Come make an impact on the communities we serve as you help us advance health optimization on a global scale. Join us to start Caring. Connecting. Growing together.

Optum AI is United Health Group's enterprise AI team. We are AI/ML scientists and engineers with deep expertise in AI/ML engineering for health care. We develop AI/ML solutions for the highest impact opportunities across United Health Group businesses including United Healthcare, Optum Financial, Optum Health, Optum Insight, and Optum Rx. In addition to transforming the health care journey through responsible AI/ML innovation, our charter also includes developing and supporting an enterprise AI/ML development platform.

Optum AI is building foundation models and agentic systems that can understand complex healthcare data, reason about clinical workflows, and safely assist clinicians and patients 're looking for a lead research scientist who wants to push the frontier of:

• Pretraining and post‑training of LLMs/SLMs (including RL, RLHF, RLAIF), and
• Agentic systems that plan, use tools, and operate in real healthcare workflows.

This role is ideal for a researcher who enjoys turning open‑ended ideas into state‑of‑the‑art models, publishing in top venues, and seeing their work deployed to improve health outcomes for millions of people.

You'll enjoy the flexibility to work remotely
* from anywhere within the U.S. as you take on some tough challenges. For all hires in the Minneapolis or Washington, D.C. area, you will be required to work in the office a minimum of four days per week.

• Lead research on pretraining and post‑training of language models
  • Design and train domain‑specialized LLMs and smaller language models (SLMs) for healthcare applications
  • Explore novel pretraining objectives, architectures, and data curation strategies
  • Develop post‑training pipelines (RL, RLHF, RLAIF) to align models with clinical best practices, safety guidelines, and user preferences
  • Develop RL/RLHF/RLAIF methods at production scale
  • Design reward models and feedback collection strategies with clinicians and domain experts
  • Implement and evaluate RL‑based fine‑tuning of foundation models using human and AI feedback
  • Work with platform teams to run large‑scale, distributed experiments on modern GPU/cloud infrastructure
• Build and study agentic systems for healthcare
  • Design agent architectures that can plan, call tools/APIs, interact with retrieval systems (eg, RAG), and handle multi‑step clinical workflows
  • Investigate memory, planning, and tool‑use strategies to make agents reliable, controllable, and debuggable
  • Define and evaluate benchmarks for agent performance, robustness, and safety in healthcare contexts
• Drive research to production and publish your work
  • Own the end‑to‑end lifecycle of research projects: problem formulation, literature review, experimentation, evaluation, and iteration
  • Collaborate with product teams to translate research into deployed systems that support clinicians, care managers, and patients
  • Publish findings in top‑tier AI and ML venues (eg, NeurIPS, ICML, ICLR, ACL, EMNLP, KDD, MLHC, CHIL) and present internally to Optum AI and business leaders
  • Ensure safety, fairness, and responsible AI
  • Closely collaborate with the Responsible Use of AI (RUAI) team to embed safety, fairness, and compliance into model and agent design
  • Develop evaluation protocols, diagnostics, and documentation to support model governance and regulatory requirements
• Contribute to high‑quality engineering and collaboration
  • Build robust training and evaluation pipelines using Python, PyTorch/Tensor Flow, and modern ML tooling
  • Follow best practices for software development (tests, packaging, Git Hub, code review)
  • Communicate clearly with…