Backend Developer Needed: Structured AI Clinical Reasoning API (PROMIS / Healthcare)
We are building an early-stage healthcare software product focused on structured clinical reasoning, not chatbots.
The goal is a small, reliable backend service that accepts standardized inputs (PROMIS scores + minimal clinical context) and returns a strictly validated JSON output representing clinical reasoning for non-specific low back pain in a Workers’ Compensation context.
This is not a full app.
It is not UI-heavy.
It is a precision backend problem.
If you enjoy clean APIs, schemas, validation, and test-driven development, this will be a good fit.
See the following links to read more about PROMIS:
https://commonfund.nih.gov/promis
https://www.promishealth.org/57461-2/
https://www.sralab.org/rehabilitation-measures/promis-ability-participate-social-roles-and-activities
What you will build (V0)
A single POST endpoint: /generate-plan
Input: JSON (PROMIS Physical Function / Pain Interference + clinical + WC context)
Output: JSON that must validate against a provided JSON Schema
LLM call with:
strict structured output
retry-on-failure logic
low temperature
Local evidence card store (simple JSON or DB)
Automated tests (10 scenarios provided)
Dockerfile + README
No EMR integrations.
No frontend required (unless you strongly prefer a minimal one).
Tech Stack (flexible)
Backend: Node.js (Express) or Python (FastAPI)
Schema validation: AJV / pydantic
LLM: OpenAI or compatible API
Tests: Jest / Pytest
Deployment: Render, Railway, Fly.io, or Docker-local
Non-negotiables
Strict JSON schema validation (no free-text output)
Clean code + repo access from day 1
Automated tests
Clear README
No PHI storage or logging
Willingness to work from a detailed spec
Nice to have
Experience with LLM structured outputs
Healthcare or regulated-domain experience
Opinionated about validation and failure handling
How to apply (important)
In your first message, please include:
A short description of a backend API you’ve built with strict validation
Which stack you’d use here and why
How you would handle LLM output that fails schema validation
Your availability over the next 2–3 weeks
Applications that don’t address these points will not be considered.
Budget
Fixed-scope, milestone-based.
Looking for someone efficient and pragmatic rather than flashy.
Apply tot his job
Apply To this Job