Staff AI Scientist

Ōura • San Francisco, CA

7d • Hybrid

About The Position

The Health Intelligence team is at the forefront of integrating modern AI and LLMs into the Oura experience, transforming how members interact with and learn from their data. We are building a next-generation AI-powered personalization platform at the intersection of classical recommendation systems and modern AI. The retrieval and ranking engine draws on deep collaborative signals, graph-based user representations, and longitudinal health data. The serving layer increasingly runs through LLMs, which translates what the engine knows into language that is relevant, personalized, and safe. Getting that handoff right — ensuring the GenAI layer serves the model's intent rather than overriding it — is one of the defining technical challenges of this role.
As a Staff AI Scientist, you will set the technical direction for personalization at Oura. You will be hands-on in building, deploying, and iterating on production systems, and you will hold a high bar for the velocity at which the team moves from hypothesis to live experiment to learning. You will work across the full stack of the personalization system — from representation and retrieval through ranking, generation, and evaluation — and you will be the connective tissue between data science, engineering, product, and design. This is a high-visibility role for someone who thinks in systems, ships with urgency, and wants to build something that compounds in value over weeks and months.
This is a US Hybrid role based in San Francisco, CA. Candidates should be able to work onsite in our San Francisco office 2-3 days per week.

Requirements

• 8+ years of experience in applied machine learning or AI research, with a demonstrated expertise in recommendation systems, personalization, or retrieval. A graduate degree (MS or PhD) in a relevant quantitative field such as Computer Science, Statistics, or a related discipline is strongly preferred.
• Hands-on experience across retrieval, ranking, and recommendation system design (including collaborative filtering, embedding-based approaches, graph networks, or related methods), and a track record of shipping these into real production systems in a robust experimentation framework, not just offline analyses or research prototypes.
• Comfort working closely with server and app engineers on model serving, pipeline architecture, and deployment infrastructure — and an instinct for where to cut scope to ship faster without compromising the science.
• Practical experience integrating recommendation or retrieval signals with LLM-powered generation, including work on grounding, constrained decoding, prompt design, or evaluation frameworks that assess whether the generation layer preserved upstream intent.
• Demonstrated ability to design lightweight experiments and evaluations that generate signal quickly, such as shadow testing, staged rollouts, and proxy metrics that responsibly accelerate the learning loop without waiting on long A/B cycles.
• Experience framing personalization problems, modeling user trajectories, and working with stateful or sequential data.
• Solid exposure to causal methods (uplift modeling, treatment effect estimation, counterfactual evaluation) and experiment design, with the ability to interpret results with appropriate caution and communicate uncertainty clearly.
• Evidence of operating beyond individual contributions: influencing technical direction, mentoring others, shaping team practices, or leading cross-functional scientific initiatives.
• Strong ability to explain complex systems, trade-offs, and uncertainty to both technical and non-technical audiences, and to operate effectively in a fast-moving, ambiguous domain.
• Strong proficiency in Python, including data analysis and modeling, as well as experience with modern data tooling in collaboration with data and engineering partners.

Nice To Haves

• Experience designing personalization specifically for consumer behavior change or health outcomes, where the goal is longitudinal impact rather than short-term engagement.
• Familiarity with health, wearables, or digital therapeutics domains, and genuine interest in how personalization compounds over a member's lifetime.
• Deep experience with AI / LLM-backed products and evaluation workflows, such as LLM-as-judge, rubric-based evaluation, safety/red-teaming, and offline vs. online assessment of model quality, latency, and cost.
• Comfort working outside “core” hours across time zones with distributed, cross-functional teams.

Responsibilities

• Define the personalization tech strategy: Set the research and modeling agenda for how Oura represents users, retrieves relevant content and interventions, and ranks and delivers them across surfaces. Identify where classical approaches (collaborative filtering, graph networks, similarity-based retrieval) are the right foundation and where newer methods add genuine value. Influence roadmap and technical direction across partner teams.
• Own user representation and retrieval: Build and maintain rich, longitudinal user state representations that span physiology, behavior, goals, preferences, and context. Design retrieval systems that operate over these representations to surface the right content, interventions, or guidance at the right moment.
• Architect a modern personalization serving interface: Define how personalization signals from the retrieval and ranking engine are passed to and preserved by the LLM serving layer. Develop grounding and constraints that prevent the serving layer from drifting away from what the ranking engine decided, ensuring GenAI serves as a personalization-aware delivery mechanism.
• Drive evaluation rigor: Design measurement frameworks that assess the full chain: retrieval quality, ranking calibration, and whether the GenAI serving layer preserved intent and personalization signal. But evaluation only matters if it moves fast enough to inform the next decision — you will build lightweight offline evals and shadow-mode testing infrastructure that let the team iterate quickly without waiting for long A/B cycles. Establish rubrics and tooling others can use and reuse.
• Apply causal reasoning to understand what works: Own the causal and counterfactual reasoning necessary to distinguish personalization effects from confounding. Design and analyze experiments that measure genuine impact on behavior and health, not just engagement.
• Mentor and raise the bar As a Staff scientist, you are expected to grow the people around you by providing technical mentorship to scientists and engineers — shaping team norms around experimentation and evaluation, and helping define what good looks like for personalization science at Oura.
• Collaborate and communicate across functions Partner with engineering, science, product, and design across the Health Intelligence team to shape how personalization integrates into the broader member experience. Communicate trade-offs, uncertainty, and modeling assumptions clearly to technical and non-technical stakeholders across the US and EU.

Benefits

• Competitive salary and equity packages
• Health, dental, vision insurance, and mental health resources
• An Oura Ring of your own plus employee discounts for friends & family
• 20 days of paid time off plus 13 paid holidays plus 8 days of flexible wellness time off
• Paid sick leave and parental leave

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level