Machine Learning Engineer (MLOps & Data)
About Surgeon Decision Intelligence (SDI)SDI is a healthcare AI startup building surgical decision-support tools for spine surgery. Founded bypracticing spine surgeons and academic faculty, we combine deep clinical expertise with modernmachine learning to improve surgical planning and patient outcomes. Our platform spans medicalimaging pipelines, clinical outcomes data, and ML models that operate on both.
The role:
We have working models: code that takes in an MRI series and produces measurements surgeons careabout. We need an engineer to turn them into production systems that are tested, versioned, deployedand monitored, with every output traceable to the model, code and data that produced it. This is anMLOps-focused role: reproducibility, versioning, deployment and monitoring of our models are the coreof the job. An implementation partner is currently standing up our core platform: Databricks, MLflow, ourdatabases and model serving. You'll work alongside them during the build, take ownership when theyhand it over, and be the person who gets our models onto it and keeps them running. You'll also helpgrow and connect the data our research depends on. This is a hands-on engineering role on a small team. You'll work closely with our Principal AI/ML Engineeron how models are built, tracked, released and served. What you'll work on in your first year Take our models from prototype to production. Refactor MVP model code, starting with our MRIimaging pipeline, into tested, typed, reproducible Python packages. Find out what's slow and make itfast. Make every result reproducible and traceable. Register models in MLflow with pinned code, environment and data versions. Build batch and online inference on Databricks so that every storedoutput records the model that produced it. Define how a new model is validated, promoted androlled back, and write those procedures down. Build the path from research code to production. Set up project templates, CI, code reviewand a clear path from "my model works" to "my model is registered and deployable" for everyone onthe team who writes model code. Grow and connect our research data. Our core research datasets already exist. You'll help shapehow they need to grow and change as we do, and connect our data sources (imaging, annotations, scanner and protocol metadata, clinical outcomes) so researchers can find, query and use them. That includes designing schemas, versioning the datasets our models are trained and evaluated on, and writing loaders that handle the edge cases real-world data throws at you. Own the platform. Once our partner hands it over, you'll run it day to day: CI/CD, access, cost andupgrades. You'll also set up monitoring for the services and for data and model drift, so we find outabout problems before our users do.
What we're looking for:
2+ years of professional software engineering, mostly in Python, writing tested code that runs inproduction.
You've taken at least one ML system from prototype to production yourself and kept it runningafterward, so you know what breaks.
A working understanding of how models are trained and evaluated, enough to reproduce a trainingrun: seeds, nondeterministic GPU operations, data splits, preprocessing versions. Experience with
PyTorch.
Hands-on MLOps: experiment tracking or a model registry (MLflow or similar), deploying modelinference, and CI/CD.Data engineering: you've built pipelines that produce datasets other people train on. Relationaldatabase design in PostgreSQL or similar: schemas, migrations, constraints, and knowing when notto normalize.
You've reviewed other people's code and helped less experienced developers write better softwarewithout slowing them down.
You've worked at a startup or on a small team where you owned work end to end.
You can take a loosely defined problem, ask the right questions, and come back with a workingsolution without needing step-by-step direction.
Nice to have:
Databricks (strongly preferred): Unity Catalog, Delta Lake, Spark, Model Serving, Workflows.
Medical imaging experience: DICOM, MRI or CT pipelines, de-identification.
Longitudinal clinical or outcomes data, including follow-up and censored time-to-event data.
Regulated healthcare software: HIPAA and handling patient data, FDA software as a medical device, design controls.
Azure (blob storage, Entra ID) or another major cloud.
Building and deploying APIs (FastAPI or similar).Storing and compressing vectors: pgvector, quantization, dimensionality reduction.
Speeding up numerical Python code (NumPy, PyTorch, GPUs).Side projects you've built and shipped on your own.
What this role is not
Not primarily a research or modeling role. Your main job is making our models reproducible andreliable in production, and building the data they're trained on. You'll need to understand ourmodels, and because we're a small team you may contribute to research from time to time, but itwon't be the core of the role.
Not building a platform from scratch, or tuning one component of a large system. Our coreinfrastructure is being stood up now. Your job is to own it and put our work on it, across the stack.
Location: Remote to start, transitioning to in person in New York City
Employment type: Full-time
Compensation: $140,000–$175,000 base salary + equity + benefits
Surgeon Decision Intelligence provides equal employment opportunities to all employees and applicants without regard to race, color, religion, age, sex, national origin, disability status, genetics, protected veteran status, sexual orientation, gender identity or expression, or any other characteristic protected by federal, state or local laws, and prohibits discrimination and harassment of any type.