Skip to content

medplum-provider-jev

Medplum demo where a bot uses Jev to compare a provider's visit note with a discharge summary and flag agreement or potential conflict.

Appprototype

Overview

Medplum Provider + Jev is a guided demo in the Medplum Provider App. A tester plays a primary care provider at a post-discharge follow-up visit. When they finish the visit, a Medplum Bot compares their free-text note with an outside hospital discharge summary using Jev, and a review card shows one of agreement, potential conflict or insufficient information. The provider can edit the note, sign with a documented reason, or create a reconciliation task. The bot also looks up medication ingredients via RxNorm codes.

The README stresses that it uses synthetic data only and that one synthetic run is not clinical validation; it should be validated with clinicians before adapting to real workflows. A separate guide covers self-hosting an open 27B model on a private Modal GPU instead of hosted Jev for proper HIPAA compliance, and an article explains the workflow, model choices, results and limits. Setup requires a Medplum project with Bots enabled, Medplum terminology including RxNorm, Node 22.18+ and a TypeSafe key.

What the decision model decides

TypeQuestionAnswers
ChoiceDoes the provider's note agree with the discharge summary?agreement · potential_conflict · insufficient_information

How it works

  1. Tester completes a synthetic follow-up visit note
  2. Medplum Bot compares the note with the discharge summary via Jev
  3. Medication ingredients are resolved via RxNorm
  4. Review card shows agreement, potential conflict or insufficient information
  5. Provider edits, signs with a reason, or creates a reconciliation task

Strengths

  • Integrated into a real EHR-style workflow
  • Human stays in control of resolution
  • Self-hosting path for HIPAA needs

Caveats

  • Synthetic data only; not clinically validated
  • Requires Medplum Bots and terminology setup
  • Hosted Jev not suitable for PHI without proper agreements

Getting started

npm --prefix provider ci --ignore-scripts, log in with the Medplum CLI, run the setup script, then npm --prefix provider run dev and open http://localhost:3001.

At a glance

Type
App
Status
prototype
Language
TypeScript
Runtime
Medplum Provider App and Bot
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Whether the visit note agrees with, potentially conflicts with, or lacks information relative to the discharge summary.

Can I use real patient data?

No; the demo uses synthetic data and the README describes self-hosting for HIPAA needs.

Who resolves conflicts?

The provider, by editing the note, documenting a reason or creating a task.

Related projects

  • TLA+-verified consensus kernel around Jev that may escalate but never returns a confident wrong verdict, tested on 1,680 pharmacy cases.

    ProjectresearchRust
  • Spanish oncology decision-support prototype that walks ESMO guideline trees and uses Jev only for judgment calls.

    ProjectresearchPython
  • Research implementation of Jev-style typed decisions over clinical evidence: supported, contradicted or unresolved.

    ProjectresearchPython
  • Pre-registered benchmark of Jev as a medical hallucination check on MedHallu, versus fast LLMs, on accuracy, latency and cost.

    Benchmarkresearch
  • Published Jev client and claim check from the Fake/Real evidence-review site, where Jev filters sources and judges each claim.

    ProjectproductionTypeScript
  • SRT subtitle translator that translates with any OpenRouter LLM and asks Jev a review Noul per line to flag defects before export.

    ProjectbetaPython