Job description

THE ROLE Truemed enables health and wellness merchants to accept $150+ billion in HSA/FSA funds. At the center of that is determining whether products qualify for pretax spending - a classification problem with regulatory, financial, and clinical consequences at scale. We solve this using cutting-edge ML technologies that enable us to scale the number of merchants and product categories we can support while maintaining industry-leading accuracy requirements. As a Senior ML Engineer, you will run the experiments that decide how that system improves, and ship the winners to production. WHAT YOU'LL DO - Run the experiments that move classification accuracy: form a hypothesis, test it against labeled data, and ship what wins - Design and iterate on classifiers that evaluate products against eligibility rules, across prompting, retrieval, fine-tuning, and multi-model approaches - Dig into misclassifications to find failure modes, and turn each one into the next experiment - Build and extend the evaluation sets and metrics that tell us when a model is better, including how we label ground truth - Track new models and research, and test the ones that could change our accuracy or cost curve - Extend classification coverage to new product categories as Truemed expands - Find new applications for our models and data beyond product classification, and prototype the ones worth pursuing - Collaborate with clinical ops and legal teams to turn eligibility standards into labeled data and testable system behavior - Write up results so the team makes model decisions from evidence, and help raise AI fluency across engineering WHAT YOU BRING - 5+ years of engineering or applied research experience, with at least 2 years of hands-on work with LLMs or other ML models on real problems - A track record of rigorous experiments: clear hypotheses, sound baselines, ablations, and conclusions you would defend - You thrive and enjoy looking at data with your eyes for hours and consider it part of the workflow in any ML work - Comfort designing evaluation methodology for AI systems - defining what "correct" means and measuring it - Hands-on experience with prompting, fine-tuning, retrieval, or agent design on foundation models (OpenAI, Anthropic, open-weight models) - Fluency in Python and enough engineering depth to take an experiment to production - Senior-level ownership: you've taken an ambiguous ML problem from first experiment to production on your own - Comfort with ambiguity - the right approach isn't always known upfront and has to be discovered through experimentation - Public evidence of how you run experiments: a paper, open-source ML work, or a technical write-up of results you would defend - Bonus: experience with healthcare or regulatory classification, or multi-model orchestration ABOUT TRUEMED Our mission is to help people take control of their health. Today, 95% of all medical costs go to treating illness. Our goal is to empower people to invest in healthy habits today rather than treating illness in the future. We do this by supporting these 5 habits that research shows are tied to significantly improved health outcomes: 1. Eating healthy food 2. Getting an adequate quantity and quality of sleep 3. Increasing movement and exercise and reducing sedentary behavior 4. Reducing chronic stress 5. Increasing sunlight and minimizing environmental toxins Today Truemed enables patients to use pretax HSA/FSA funds to buy healthy food, exercise equipment, and supplements. We support brands like Peloton, Eight Sleep, Momentous, and hundreds of others. We're building the infrastructure that allows health and wellness merchants to accept the $150+ billion tied up in HSA and FSA accounts. By allowing consumers to spend these pre-tax dollars on goods that make them healthier, we'll incentivize consumers to invest in their health and begin to reverse the explosion in chronic illness that's plagued the US for the last 50 years. COMPENSATION AND BENEFITS - Comp and equity in the top decile for Series A startups. This position's pay band is $205k to $250k plus stock options - Comprehensive medical, dental, and vision insurance with employer-sponsored HSA contributions - 401(k), unlimited PTO, remote work - Coworking stipend, L&D stipend - 2x/year company offsites and 2x/year team offsites

Requirements

Must-haves

Run the experiments that move classification accuracy: form a hypothesis, test it against labeled data, and ship what wins
Design and iterate on classifiers that evaluate products against eligibility rules, across prompting, retrieval, fine-tuning, and multi-model approaches
Dig into misclassifications to find failure modes, and turn each one into the next experiment
Build and extend the evaluation sets and metrics that tell us when a model is better, including how we label ground truth
Track new models and research, and test the ones that could change our accuracy or cost curve
Extend classification coverage to new product categories as Truemed expands
Find new applications for our models and data beyond product classification, and prototype the ones worth pursuing
Collaborate with clinical ops and legal teams to turn eligibility standards into labeled data and testable system behavior
Write up results so the team makes model decisions from evidence, and help raise AI fluency across engineering
5+ years of engineering or applied research experience, with at least 2 years of hands-on work with LLMs or other ML models on real problems
A track record of rigorous experiments: clear hypotheses, sound baselines, ablations, and conclusions you would defend
You thrive and enjoy looking at data with your eyes for hours and consider it part of the workflow in any ML work
Comfort designing evaluation methodology for AI systems - defining what "correct" means and measuring it
Hands-on experience with prompting, fine-tuning, retrieval, or agent design on foundation models (OpenAI, Anthropic, open-weight models)
Fluency in Python and enough engineering depth to take an experiment to production
Senior-level ownership: you've taken an ambiguous ML problem from first experiment to production on your own
Comfort with ambiguity - the right approach isn't always known upfront and has to be discovered through experimentation
Public evidence of how you run experiments: a paper, open-source ML work, or a technical write-up of results you would defend

Nice-to-haves

Bonus: experience with healthcare or regulatory classification, or multi-model orchestration

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlaipythongo

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Dental

Vision

401(K)

Pto

Wellness