Machine Learning Intern

About stēle
We are building Stele Fusion, the first wearable that reads both brain and body. It sits in the ear all day and through the night, and it is a genuine pair of earbuds while it does it. Out of that one object come sleep, alertness, focus, recovery, stress, and the early signals of getting sick. The app is where all of it becomes something a person can actually use.
We are a luxury brand built on technology rather than a technology company. Our moat is three things: algorithms, design, and brand. This seat owns a large piece of the first.
The role
You will turn health data into insight that people can trust and that our AI can use. You will take cleaned datasets of brain and body signals and find what is really in them. Each finding becomes a validated, structured insight stored in the Stele AI system. The app draws on those insights when it explains someone's night or answers their question.
What you'll own
Turning data into insight:
Sleep and physiological datasets, including ear-EEG recorded alongside clinical sleep studies, plus heart rate, HRV, temperature, and motion
Models that pull real signal out of noisy wearable data, starting with sleep: staging, sleep quality, and what disrupts a night
Measuring what each sensor actually adds, so hardware and product decisions rest on evidence
Benchmarking every result against gold-standard labels, and saying plainly where the ceiling is
Making it usable by the AI:
Turning each validated finding into a structured insight: what it means, the evidence behind it, how confident we are, and when it applies
Designing how those insights are stored so the AI system can find them and use them accurately
Working with the app team so an insight reaches a person clearly enough to act on
What we look for
Strong math and statistics: you know why a model works and where it breaks, beyond calling a library
Python and the standard stack: NumPy, pandas, scikit-learn, and PyTorch or similar
Experience with time-series or signal data. Biosignals, sensors, audio, and financial data all count.
Discipline about overfitting and leakage, including evaluating across people rather than across samples
The ability to explain a result in one paragraph to someone outside ML
What this role is not
Pure research. A result counts once it is validated, stored, and usable in the product.
Data cleaning. You start from prepared data and own the step from data to meaning.
A place for inflated claims. We would rather ship a modest insight that holds than an impressive one that falls apart.
Strong signals
EEG, sleep science, or wearable data in your background
Experience with retrieval, embeddings, or knowledge systems that feed an LLM
A paper, a competition result, or a project where you caught a mistake others missed
How we work
Small team and direct access to the founders. The ML team is small enough that your work shows up in the product and in front of investors. You will own your problem from dataset to shipped insight.
The details
Fall 2026, around 20 hours a week, remote within the United States. Unpaid to start, which is true for everyone here right now, and we will revisit that when our funding round closes. We are happy to work with your school if you want this to count for academic credit.
How to apply
Send your resume and one project you are proud of: code, a paper, or a writeup. Tell us what the data looked like, what you found, and one thing you got wrong along the way.

Work arrangement
Yes