About the role
About Sprinter Health At Sprinter Health, our mission is reimagining how people access care by bringing it directly to their homes. Nearly 30% of patients in the U.S. skip preventive or chronic care simply because they can’t get to a doctor’s office. For many, the ER becomes their first touchpoint with the healthcare system, driving over $300B in avoidable costs every year. By using the same technologies that power leading marketplace and last-mile platforms, we deliver care where people are, especially those who need it most. So far, we’ve supported more than 2 million patients across 22 states, completed 130,000+ in-home visits, and maintained a 92 NPS. Our team of clinicians, technologists, and operators has raised over $125M from investors like a16z, General Catalyst, GV, and Accel and enjoys multi-year runway.
About
the Role We’re looking for an Applied Scientist, AI to turn messy, high-stakes healthcare problems into machine learning models and AI systems that improve access to care and help Sprinter operate more effectively. This role sits at the intersection of research, product, engineering, and clinical operations. You’ll take ambiguous product and operational problems and turn them into well-scoped prediction, ranking, optimization, NLP, or LLM-based tasks. You’ll build strong baselines, design honest evaluations, run careful error analysis, and iterate toward models that can improve real-world outcomes. The right person for this role combines scientific rigor with a deployment-oriented mindset. You should care deeply about evaluation, leakage, bias, confounding, and whether offline results actually translate into production impact. You should also be able to partner closely with ML engineering to productionize models, work with clinicians and subject-matter experts to validate assumptions, and explain model behavior, uncertainty, and limitations clearly to product and leadership. This role is ideal for a scientist-engineer who can move fluidly between data exploration, modeling, experimentation, error analysis, stakeholder partnership, and production handoff. Office Location We are a hybrid company based in the Bay Area with offices in both San Francisco and Menlo Park. We operate on a hybrid schedule, working from the office Monday through Thursday, with Fridays designated as work-from-anywhere days. We care deeply about work-life balance and are happy to provide flexibility when life happens. We ask that employees be in the office Monday through Thursday to collaborate with their teams while maintaining flexibility where it matters most. Lunch is provided every day, and the entire team takes an hour to eat together. It’s one of the ways we stay connected outside of meetings. You’ll usually find us playing a board game before getting back to work. What you will do Turn ambiguous healthcare, product, and operational problems into well-posed ML, AI, ranking, optimization, NLP, or LLM-based tasks Build strong baselines and improve on them efficiently using the right modeling approach for the problem Develop models across traditional ML, deep learning, NLP, and LLM-based approaches where appropriate Design offline and online evaluations that are honest, measurable, and predictive of real-world impact Choose metrics suited to imbalanced, delayed, noisy, and partially observed healthcare outcomes Run careful error analysis and use it to improve model quality, product fit, and operational usefulness Identify label leakage, selection bias, confounding, and other data artifacts before they reach production Explore messy real-world data, assess label quality, and determine whether a problem is ready for modeling Partner with ML engineering to productionize models reliably and define what production-readiness requires Work with clinical stakeholders and subject-matter experts to validate assumptions, review model errors, and understand edge cases Explain model tradeoffs, uncertainty, limitations, and expected impact clearly to product, operations, clinical, and leadership teams Write experiment docs, summarize findings, and help teams make informed decisions about when and how to deploy AI systems Pressure-test whether results are real, robust, and useful before recommending production use What you have done Built, evaluated, and iterated on machine learning or AI models for real-world use cases Turned ambiguous business, product, clinical, or operational problems into measurable modeling tasks Designed rigorous offline evaluations, experiments, or analyses that informed production or product decisions Worked with messy real-world datasets where labels, outcomes, and causal relationships are imperfect Used statistical reasoning, experimental design, and error analysis to understand model performance Built models using Python and standard ML or AI tooling such as PyTorch, scikit-learn, NumPy, pandas, Polars, Hugging Face, Matplotlib, or similar Compared modeling approaches and made pragmatic decisions about when to use traditional ML, LLMs, heuristics, or simpler baselines Communicated model performance, limitations, tradeoffs, and uncertainty to technical and non-technical stakeholders Partnered with engineering, product, data, operations, clinical, or domain experts to move models closer to production impact Operated with enough engineering depth to run experiments end to end and self-serve deployments or production handoffs when needed Used AI coding assistants such as Claude Code, Cursor, or similar tools as part of your development workflow What gives you an edge You have an MS or PhD in computer science, statistics, machine learning, applied math, operations research, biomedical informatics, epidemiology, or a related quantitative field You have exceptional applied experience that substitutes for formal graduate training You have depth in LLMs, ranking, NLP, uncertainty quantification, causal inference, optimization, or healthcare AI You’ve shipped models that reached production and had measurable real-world impact You’ve worked with healthcare data such as claims, EHR, clinical notes, scheduling, utilization, quality, risk, or patient engagement data You have experience working with PHI, HIPAA-aware systems, or other sensitive regulated data You know when traditional ML approaches are likely to outperform LLMs, and when LLMs are the right tool You have experience collaborating with clinicians, clinical operations teams, or other high-stakes domain experts You’ve worked in a startup or fast-moving applied environment where ambiguity, speed, and rigor all mattered What makes you successful You understand how ML models work under the hood and can explain them clearly to non-technical stakeholders You focus relentlessly on impact and know that the simplest model is often the best one You treat evaluation as one of the most important parts of model development You notice when a metric is misleading, incomplete, or disconnected from real-world outcomes You catch leakage, bias, and confounding that others miss You move fluidly between modeling, error analysis, stakeholder partnership, and production handoff You can hand a model to engineering and explain its limits to a clinician with equal clarity You are comfortable with ambiguity and can adapt modeling approaches to problems that do not come with a playbook You balance scientific rigor with the practical need to ship useful systems Day to Day In this role, you might spend your time: Exploring data and labels for a new healthcare or operational problem Turning an ambiguous product question into a measurable modeling task Building and comparing models, then running error analysis Reviewing misclassified or low-confidence cases with a clinical subject-matter expert Designing an offline evaluation that is more likely to predict online or real-world success Partnering with ML engineering to prepare a model for deployment Writing an experiment doc and presenting findings to product and leadership Pressure-testing whether a result is real or an artifact of the data Comparing a simple baseline, traditional ML model, and LLM-based approach to determine what is most useful Investigating why model performance differs across populations, workflows, labels, or operational contexts The Interview Process We aim to complete the interview process within 2–3 weeks. It will usually consist of: Recruiter Screen: Background fit, motivation, and