Applied Scientist, Sprinter Health

• This role is ideal for a scientist-engineer who can move fluidly between data exploration, modeling, experimentation, error analysis, stakeholder partnership, and production handoff
• 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

Desirable

• 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
• 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 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
• 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

• We aim to complete the interview process within 2–3 weeks. It will usually consist of:
• Recruiter Screen: Background fit, motivation, and compensation alignment
• Hiring Manager Interview: Applied science experience, modeling depth, and healthcare/product orientation
• Hands-on Technical Assessment: Practical modeling, evaluation, error analysis, and scientific judgment
• Onsite Interview: Technical case study, research or project presentation, behavioral interview, and lunch with the team
• References: Validation of performance, judgment, and working style