About Ultralight

Ultralight is the AI-native operating system for medical practices in preventive medicine (longevity, functional, integrative). Clinical software is broken: physicians spend more time fighting their tools than caring for patients, and patients have no visibility into their own health data. We're building the system that fixes both sides.

Ultralight isn't a better EHR with AI bolted on. It's an agentic harness at the center of a medical practice: deep context on every patient (labs, documents, wearables, imaging, notes in a unified record) and AI that can act in the clinical environment (ordering, parsing, summarizing). A solo practitioner with Ultralight should have the clinical leverage of an entire care organization.

We're seed stage with 75 practices on the platform, a small team that ships fast, and a product surface that spans a full clinician web platform and a React Native patient mobile app.

Role Overview

Every new practice on Ultralight brings years of patient history from a prior EHR. Getting that data in cleanly, safely, and repeatably is what turns a launch into a win or a drag. You own that.

This isn't customer success with a keyboard. And it isn't an implementation engineer who takes tickets from a platform team. The migration stack is yours end to end: deterministic handlers, an agentic extractor, FHIR upload, quarantine review, async document parsing. Code, CLI, control plane, customer conversation.

Today the stack works, but much of it still lives in TypeScript scripts and runbooks. Your job is to turn it into a product that future migrations flow through without hero-mode effort.

What You'll Be Working On

• You own customer migrations from scoping through sign-off. You run the dry run, you explain what was extracted and what wasn't, you make the call on approve, reject, or retry
• You extend the migration agent: handlers for known export formats, the organizer, the extractor, the uploader, and the prompt and skill layer the agent leans on
• You convert successful agentic discovery into deterministic handlers. The thing that works once should work forever
• You productize the operator surface: approve and reject flows in the admin UI, dataset discoverability, telemetry on parse health and pending documents. The next migration shouldn't require a terminal
• You debug production-adjacent failures across organizer, extractor, upload, quarantine, and document parsing. The whole stack is yours to reason about
• You work with platform engineering on the Medplum and FHIR edges, and with our Practice Success team on UAT and rollout order

Who We're Looking For

• Strong TypeScript and internal-platform engineering. You've built systems other engineers depend on, not just features
• You've worked on LLM tool-use or agent workflows in production. You know where they're sharp and where they're soft
• Comfort with CLI-heavy operational work. Long-running jobs, resumable state, logs you actually read
• You can debug across layers: an unexpected parse result, a silent upload failure, a downstream async job that didn't fire