Imported from jeremylongshore/tons-of-skills-marketplace (skills/.curated/abridge-local-dev-loop/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill abridge-local-dev-loop. Copyright stays with the author (MIT).

Abridge Local Dev Loop

Overview

Local development workflow for Abridge clinical AI integrations. Uses a local HAPI FHIR server for EHR simulation, synthetic patient data from Synthea, and Abridge sandbox APIs. Never use real PHI in development.

Prerequisites

• Completed abridge-install-auth setup
• Docker installed (for local FHIR server)
• Node.js 18+ with TypeScript
• Abridge sandbox credentials

Instructions

Step 1: Start Local FHIR Server

# Run HAPI FHIR R4 server locally
docker run -d --name hapi-fhir \
  -p 8080:8080 \
  -e hapi.fhir.default_encoding=json \
  hapiproject/hapi:latest

# Verify
curl -s http://localhost:8080/fhir/metadata | jq '.fhirVersion'
# → "4.0.1"

# Seed synthetic patient
curl -X POST http://localhost:8080/fhir/Patient \
  -H "Content-Type: application/fhir+json" \
  -d '{"resourceType":"Patient","name":[{"given":["Jane"],"family":"Doe"}],"birthDate":"1985-03-15","gender":"female"}'

Step 2: Create Dev Environment Configuration

// src/config/dev.ts
interface DevConfig {
  abridge: { baseUrl: string; clientSecret: string; orgId: string };
  fhir: { baseUrl: string };
  fixtures: { transcriptsDir: string };
}

const devConfig: DevConfig = {
  abridge: {
    baseUrl: process.env.ABRIDGE_SANDBOX_URL || 'https://sandbox.api.abridge.com/v1',
    clientSecret: process.env.ABRIDGE_CLIENT_SECRET!,
    orgId: process.env.ABRIDGE_ORG_ID!,
  },
  fhir: { baseUrl: 'http://localhost:8080/fhir' },
  fixtures: { transcriptsDir: './fixtures/transcripts' },
};

export default devConfig;

Step 3: Build Transcript Fixtures

// fixtures/transcripts/index.ts
export const CARDIOLOGY_VISIT = {
  specialty: 'cardiology',
  segments: [
    { speaker: 'provider', text: 'I see you are here for a blood pressure follow-up.', timestamp_ms: 0 },
    { speaker: 'patient', text: 'Yes, I have been taking the lisinopril like you prescribed.', timestamp_ms: 3500 },
    { speaker: 'provider', text: 'Your BP today is 138 over 85. Better, but still elevated. Let us increase lisinopril from 10 to 20mg.', timestamp_ms: 7200 },
    { speaker: 'patient', text: 'Okay. Should I be worried?', timestamp_ms: 15000 },
    { speaker: 'provider', text: 'Not at all. Come back in four weeks.', timestamp_ms: 18000 },
  ],
};

export const DERMATOLOGY_VISIT = {
  specialty: 'dermatology',
  segments: [
    { speaker: 'provider', text: 'What brings you in today?', timestamp_ms: 0 },
    { speaker: 'patient', text: 'I have this mole on my back that has been changing color.', timestamp_ms: 2500 },
    { speaker: 'provider', text: 'How long has it been changing? Any itching or bleeding?', timestamp_ms: 5000 },
    { speaker: 'patient', text: 'About three months. It itches sometimes.', timestamp_ms: 8000 },
    { speaker: 'provider', text: 'The borders look irregular. I want to do a biopsy today.', timestamp_ms: 12000 },
  ],
};

Step 4: Dev Test Runner with Watch Mode

// src/dev/test-encounter.ts
import devConfig from '../config/dev';
import { CARDIOLOGY_VISIT } from '../../fixtures/transcripts';
import axios from 'axios';

async function runDevEncounter() {
  const api = axios.create({
    baseURL: devConfig.abridge.baseUrl,
    headers: {
      'Authorization': `Bearer ${devConfig.abridge.clientSecret}`,
      'X-Org-Id': devConfig.abridge.orgId,
    },
  });

  const { data: session } = await api.post('/encounters/sessions', {
    patient_id: 'demo-patient-001',
    provider_id: 'demo-provider-001',
    encounter_type: 'outpatient',
    specialty: CARDIOLOGY_VISIT.specialty,
    sandbox: true,
  });

  for (const seg of CARDIOLOGY_VISIT.segments) {
    await api.post(`/encounters/sessions/${session.session_id}/transcript`, seg);
  }

  await api.post(`/encounters/sessions/${session.session_id}/finalize`);

  // Poll for note
  for (let i = 0; i < 30; i++) {
    const { data } = await api.get(`/encounters/sessions/${session.session_id}/note`);
    if (data.status === 'completed') {
      console.log(JSON.stringify(data.note.sections, null, 2));
      // Push to local FHIR
      await axios.post(`${devConfig.fhir.baseUrl}/DocumentReference`, {
        resourceType: 'DocumentReference',
        status: 'current',
        content: [{ attachment: { contentType: 'text/plain', data: Buffer.from(JSON.stringify(data.note.sections)).toString('base64') } }],
      });
      console.log('Note pushed to local FHIR server');
      return;
    }
    await new Promise(r => setTimeout(r, 2000));
  }
}

runDevEncounter().catch(console.error);

Step 5: Package Scripts

{
  "scripts": {
    "dev:fhir": "docker start hapi-fhir 2>/dev/null || docker run -d --name hapi-fhir -p 8080:8080 hapiproject/hapi:latest",
    "dev:encounter": "tsx watch src/dev/test-encounter.ts",
    "dev:all": "npm run dev:fhir && npm run dev:encounter",
    "test:fixtures": "vitest run --grep 'fixture'"
  }
}

Output

• Local HAPI FHIR R4 server on port 8080
• Synthetic patient data seeded
• Specialty-specific transcript fixtures
• Watch-mode dev loop with live note generation

Examples

Start the local HAPI server, seed only the fictional patient fixture, and run the cardiology encounter against Abridge's sandbox. The successful local loop creates a sandbox session, emits completed note sections, and stores a DocumentReference in the local FHIR server. Inspect the resource through the local endpoint before changing application code. If Docker cannot bind port 8080 or the FHIR write fails validation, stop the watcher, correct the local configuration or resource shape, and rerun with the same synthetic fixture; do not connect a developer machine to production EHR data.

Error Handling

Resources

• HAPI FHIR Server
• Synthea Patient Generator

Next Steps

For reusable SDK patterns, see abridge-sdk-patterns.