The roleThis is the first Machine Learning Engineer role at Metriport. We have access to the richest clinical datasets in the country - longitudinal medical records for hundreds of millions of individuals - and we've barely scratched the surface of what can be learned from it. You'll own machine learning at Metriport end-to-end: from framing the problem with customers, to training the models, to serving them in production and keeping them honest.
This is applied ML on messy, high-dimensional, real-world healthcare data - not a research role, and not an LLM-wrapper role.
- You have deep ML fundamentals, from classical methods through deep learning. You can take a prediction problem from a linear baseline to gradient-boosted trees to a neural network - and you know when each one is the right answer.
- You've shipped models to production and lived with them: you have opinions about evaluation, monitoring, retraining, and what can break after launch.
- You're deeply experienced with ML, but still have strong eng chops: you can stand up a service, write the IaC, instrument it, and own it end-to-end. Python is home; TypeScript, AWS, and SQL aren't going to scare you.
- You're pragmatic about LLMs. You've used them where they win and you know where a smaller, cheaper, more reliable model wins instead.
- You're comfortable with messy, real-world data: sparse, inconsistent, high-dimensional, and full of surprises. Healthcare data is all of these at once.
- You're entrepreneurial-minded with an olympian-level work ethic (about half our engineering team are former founders).
- When someone scopes a project for 3 weeks, you ask "why can't it be done in 3 days?" - and you help others develop that same instinct.
- You're a hacker at heart, with a good sense of which rules should, and shouldn't, be broken.
What you'll be doingYou'll build the models, and the ML platform underneath them, that turn raw clinical data into intelligence our customers act on.
Day to day, that looks like:
- Building predictive models on clinical data at scale: risk stratification, expected utilization / prediction, care-gap detection.
- Owning the full ML lifecycle: problem framing, feature engineering over sparse and high-dimensional clinical data, training, evaluation, deployment, and monitoring for drift and degradation in production.
- Turning unstructured clinical data into structured, usable records: parsing PDFs, images, doctor notes, etc.
- Standing up our ML infrastructure: training and inference pipelines, experiment tracking, model versioning, and evals - so that every model we ship is measurable, debuggable, reliable.
- Working directly with founders, engineers, and customers to figure out which ML problems are highest-leverage to solve next.
Example projects you could own:
- Building a model that predicts expected hospitalizations for new patients from their medical history.
- Turn freeform doctors' notes and scanned documents into structured data we can add to patients' medical records.
- Build classifiers to categorize billions of clinical documents with incomplete metadata.
Requirements- 7+ years building ML systems that run against real data at scale.
- Strong grounding across the ML spectrum: regression and tree-based methods, feature engineering and dimensionality reduction, and deep learning. You've built and trained models yourself, not just orchestrated APIs.
- A track record of ML delivering measurable results in production.
- Strong software engineering fundamentals: you've built or operated large-scale backend systems on the cloud (ideally AWS) and can own training pipelines, model serving, and monitoring end-to-end.
- Strong data skills: SQL, working with large datasets, and building the pipelines that feed your models.
- You're located in San Francisco or the Bay Area (or willing to relocate).
- Bonus:
- Healthcare standards/technologies: FHIR, HIE, IHE, EHR/EMR, NPI, TEFCA, ADT, HL7, HEDIS, RAF, SNOMED, LOINC, ICD-10, etc.
- Founder experience, or being the first/only ML hire at an early-stage startup.
Benefits- Competitive equity + compensation package
- Full family Platinum health insurance, dental, and vision coverage 🦷
- 401(k) retirement plan + matching
- Flexible work from home or in-office
- Healthy lunches are complimentary when working in-office (and breakfast + dinners as needed) 🍏
- Quarterly company off-sites with the team
- MacBook provided by us
- Unlimited PTO (we work hard, but trust you to take time you need to be at your best)
Our techCore business logic in Node.js and TypeScript, with Python in data and ML workflows. AWS across the board (ECS, Lambda, SQS, SNS, Batch, etc.), infrastructure as code with CDK. Data lives in S3, PostgreSQL/Aurora, DynamoDB, Snowflake, and our FHIR server - with Athena for querying S3 and SageMaker for Analytics/ML. The ideal person for this role is a generalist who picks the best tool for the job.
