Senior ML Engineer, Healthcare AI

Medonix; United States; Remote; USD 140,000 - 230,000; Full time

Job summary

Medonix is building the AI suite that powers denial management, coding, and prior-auth automation, with evaluation grounded in real claim outcomes. You will own one or two end-to-end agent workflows—from model selection and prompting to deployment—and you’ll craft eval harnesses that catch regressions before shipping.

Day-to-day work mixes rapid experimentation with scalable infra, aiming to reduce cost per claim and latency without sacrificing accuracy, while partnering with credentialed coders

Qualifications

• 5+ years in applied ML / NLP with 2+ years shipping production LLMs.
• Strong Python and PyTorch / Transformers experience.
• Production experience with retrieval, fine-tuning, and prompt engineering at scale.

Responsibilities

• Own one or two end-to-end agent workflows—from model selection and prompting to deployment.
• Build and maintain an eval harness to catch regressions before shipping.
• Improve cost-per-claim and latency while preserving accuracy metrics.
• Design and evaluate prompts and fine-tunes with a fast feedback loop.

Skills

Applied ML / NLP

Python

PyTorch

Transformers

Eval harness design

Tools

Retrieval systems

Fine-tuning pipelines

Prompt engineering

Job description

Ship the agentic AI behind Medonix's denial management, coding, and prior-auth automations, measured against real claim outcomes, not vibes.

You will work on the AI Suite: eleven specialized agents that handle medical coding, denial management, prior auth, eligibility verification, claim status, and appeals. The agents are evaluated against actual production claim outcomes (clean-claim rate, denial-overturn rate, A/R compression), so the feedback loop is short and concrete.

What you will own

• Own one or two agent workflows end-to-end: model selection, prompting, fine-tuning, evals, deployment, and the metric they ship against.
• Build and maintain the eval harness that catches regressions before they ship to customer claims.
• Improve cost-per-claim and latency without giving up accuracy on the metrics that matter.
• Partner with the credentialed coders to translate gnarly payer-policy edge cases into eval cases.

Must-haves

• 5+ years in applied ML / NLP, with at least 2 years shipping production LLM-based systems.
• Strong Python and PyTorch / Transformers comfort.
• Experience designing eval harnesses that capture the right thing, not just the easy thing.
• Production experience with retrieval, fine-tuning, and prompt engineering at scale.

Nice-to-haves

• Prior healthcare AI experience (clinical NLP, coding automation, denial workflows).
• Experience with tool-using agents, multi-step planning, or multi-agent orchestration.
• Comfort with PHI-aware data engineering (de-identification, BAA-covered pipelines).

Day in the life

Half your day is on agent design and evaluation: reviewing the latest eval run with the coders, tuning prompts or fine-tunes against the failures, and shipping the next version. The other half is on the infrastructure that makes evaluations cheap and fast. We do not wait six weeks for an experiment.