Machine Learning Researcher, Document Understanding
In short. Machine Learning Researcher, Document Understanding at Brighterway: a full-time mid-level role, on-site in San Francisco, CA. The employer does not list a salary. Main work: Build evaluation sets and metrics from real customer feedback, and make quality measurable release to release. Skills mentioned: Python, Machine learning, LLMs, NLP. Posted Oct 22, 2026, checked on the employer's site 4 min ago.
Opens Brighterway's own application page. No sign-up on Sudo Get Job.
Summary
Our product is only as good as our ability to read a 3,000-page medical file and come out with the right chronology, the right duplicates removed, and the right facts surfaced. You will push the frontier of that capability: agentic document splitting and sorting, long-context summarization, duplicate and near-duplicate detection, OCR on hostile scans, and evaluation methods that tell us when quality drifts before customers do.
Quoted from Brighterway's description, not rewritten.
Skills mentioned
About the role
About Brighterway
Brighterway builds AI-native claims software and administration for casualty claims. We started in workers' compensation medical-legal review, where our agentic document pipeline sorts, de-duplicates and summarizes some of the most document-heavy files in the industry for national medical-evaluation networks. We are now building the claims platform on top of that foundation: intake across mail, fax, phone and email, bill review, subrogation, reserving signals, a workflow language, and state and federal reporting.
We are backed by Y Combinator and Maverick Ventures. Our customers handle protected health information every day, so we run HIPAA, SOC 2 and ISO 27001 programs. The team is small and technical and sits close to customers. Everyone ships, and everyone talks to the people who use what they ship.
The role
Our product is only as good as our ability to read a 3,000-page medical file and come out with the right chronology, the right duplicates removed, and the right facts surfaced. You will push the frontier of that capability: agentic document splitting and sorting, long-context summarization, duplicate and near-duplicate detection, OCR on hostile scans, and evaluation methods that tell us when quality drifts before customers do.
What you will do
Design and run experiments on LLM-based pipelines for classification, extraction, summarization and reasoning over medical and legal documents.
Build evaluation sets and metrics from real customer feedback, and make quality measurable release to release.
Improve cost, latency and reliability of multi-step agentic pipelines without giving up accuracy.
Work with engineering to move research results into production quickly.
Follow model releases and decide when a new model or technique is worth adopting.
What we look for
Graduate degree or equivalent research experience in machine learning or NLP, or a strong record of applied ML work.
Hands-on experience with large language models: prompting, fine-tuning, retrieval, evaluation.
Strong Python and the discipline to make experiments reproducible.
Comfort with messy real-world data and imperfect ground truth.
Nice to have
Document AI, OCR or layout analysis experience.
Healthcare or legal NLP experience.
Publications or open-source work in the area.
How we work
Small team, high trust, direct access to founders and customers.
We ship to production weekly and measure ourselves on customer outcomes.
Competitive salary and meaningful equity. Health, dental and vision coverage. Flexible time off.
Brighterway is an equal opportunity employer.
Location: San Francisco, CA (in person, Financial District office) Employment type: Full-time
Opens Brighterway's own application page. No sign-up on Sudo Get Job.
