Machine Learning Engineer

Own the models the whole company is named after: the Franchise Readiness Score, the Compatibility Index, and the calibration that keeps both honest.
Team Intelligence
Type Full-time
Location Remote — anywhere with 4 hours overlap with Eastern Time
The FRS is a number people make life-changing decisions on. That makes calibration a product requirement rather than a metric: a 78 has to mean the same thing this quarter as it did last, and has to mean the same thing for two franchisees who are not alike.
The interesting problem is not accuracy, it is fairness and drift. The score must not learn proxies for things it has no business scoring, and it must stay stable as the population it sees changes. You will spend as much time on evaluation and explanation as on modelling.
What you will do
Own the FRS and FCI models end to end — features, training, evaluation, deployment
Build the calibration and drift monitoring that proves a score still means what it meant
Make every dimension explainable: a franchisee is owed the reason, not just the number
Audit for proxy bias, and design the evaluation that would catch it
Work with the domain rules rather than around them — the score lives inside consent
What we are looking for
Solid applied ML, and judgement about when a model is the wrong answer
You have shipped a model into production and lived with its failures
You can explain a prediction to someone it affects, in their language
Rigorous about evaluation — you design the test before you like the result
Comfortable that the honest answer is sometimes 'we cannot score that'
Nice to have
Credit scoring, underwriting, admissions, or another consequential-decision domain
Fairness and explainability work you can point at
Experience calibrating a score that non-technical people act on
The stack
Python
PyTorch
FastAPI
Remote
This role is fully remote. We care about overlap, not postcode — four hours with Eastern Time is enough.
How we hire
A conversation, a piece of real work paid at our rate, and a call with the team you would join. No whiteboard puzzles, no unpaid take-homes.
#J-18808-Ljbffr