Skip to content

ai_transcript_analysis_js

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

ai_transcript_analysis_js

readme.md

AI Transcript Analysis

Real-time personality trait analysis from meeting transcripts with live visualization.

Built with RTMSManager - Zoom's JavaScript library for real-time media streaming.

Quick Start

npm install
cp .env.example .env   # Fill in your credentials
node index.js

Expose with ngrok: ngrok http 3000

What This Sample Does

This sample analyzes meeting transcripts in real-time to identify personality traits like Curiosity, Empathy, Assertiveness, Creativity, and Analytical thinking. As participants speak, the AI categorizes their phrases and accumulates trait scores over time. Results are visualized in a D3.js-powered frontend that updates live as the meeting progresses.

Prerequisites

  • Node.js v18+
  • Zoom account with RTMS enabled
  • Zoom App configured in Marketplace (with Zoom Apps SDK capabilities)
  • OpenRouter API key (for AI trait analysis)
  • ngrok for local development

Environment Variables

Variable Required Description
ZOOM_SECRET_TOKEN Yes Secret token from your Zoom app
ZOOM_CLIENT_ID Yes Client ID from your Zoom app
ZOOM_CLIENT_SECRET Yes Client secret from your Zoom app
PORT No Server port (default: 3000)
WEBHOOK_PATH No Webhook endpoint path (default: /webhook)
OPENROUTER_API_KEY Yes API key for OpenRouter
DEFAULT_OPENROUTER_MODEL No AI model to use (default: meta-llama/llama-4-scout:free)

Code Walkthrough

1. Initialize RTMSManager

const rtmsConfig = {
  logging: 'info',
  logDir: path.join(__dirname, 'logs'),
  credentials: {
    meeting: {
      clientId: process.env.ZOOM_CLIENT_ID,
      clientSecret: process.env.ZOOM_CLIENT_SECRET,
      zoomSecretToken: process.env.ZOOM_SECRET_TOKEN,
    },
  },
  mediaParams: {
    audio: {
      contentType: MEDIA_PARAMS.MEDIA_CONTENT_TYPE_RAW_AUDIO,
      sampleRate: MEDIA_PARAMS.AUDIO_SAMPLE_RATE_SR_16K,
      channel: MEDIA_PARAMS.AUDIO_CHANNEL_MONO,
      codec: MEDIA_PARAMS.MEDIA_PAYLOAD_TYPE_L16,
      dataOpt: MEDIA_PARAMS.MEDIA_DATA_OPTION_AUDIO_MIXED_STREAM,
      sendRate: 100,
    },
    transcript: {
      contentType: MEDIA_PARAMS.MEDIA_CONTENT_TYPE_TEXT,
      language: MEDIA_PARAMS.LANGUAGE_ID_ENGLISH,
    },
  }
};

await RTMSManager.init(rtmsConfig);

2. Set Up Webhook Handler

const webhookManager = new WebhookManager({
  config: {
    webhookPath: process.env.WEBHOOK_PATH || '/',
    zoomSecretToken: rtmsConfig.credentials.meeting.zoomSecretToken,
  },
  app: app
});

webhookManager.on('event', (event, payload) => {
  console.log('[Consumer] Webhook Event:', event);
  RTMSManager.handleEvent(event, payload);
});

webhookManager.setup();

3. Set Up Frontend Manager and WebSocket

const frontendManager = new FrontendManager({
  config: {
    port: appConfig.port,
    serveStaticEnabled: true,
    viewsPath: path.join(__dirname, 'public'),
    frontendWssUrl: process.env.FRONTEND_WSS_URL_TO_CONNECT_TO || '',
    frontendWssPath: '/ws'
  },
  app: app
});
frontendManager.setup();

const frontendWssManager = new FrontendWssManager({
  config: {
    frontendWssEnabled: true,
    frontendWssPath: '/ws'
  },
  server: server
});
frontendWssManager.setup();

4. Handle Transcript Events with Trait Analysis

RTMSManager.on('transcript', async ({ text, userId, userName, timestamp, meetingId, streamId, productType }) => {
  console.log('Transcript received:', text);

  const { current: traitsCount, total } = await extractAndAccumulateTraits(text);

  console.log("Trait Counts (this message):", traitsCount);
  console.log("Accumulated Trait Totals:", total);

  frontendWssManager.broadcastToFrontendClients({
    type: 'transcript',
    content: text,
    user: userName,
    timestamp: Date.now(),
    traits: total
  });
});

5. Start the Server

await RTMSManager.start();

server.listen(appConfig.port, () => {
  console.log(`Server running at http://localhost:${appConfig.port}`);
  console.log(`Frontend WebSocket available at ws://localhost:${appConfig.port}/ws`);
});

Key Files

File Purpose
index.js Main entry point - initializes RTMSManager, managers, and server
chatWithOpenrouterForTraits.js AI trait extraction and accumulation logic
public/index.ejs Frontend UI template with D3.js visualization
public/lib/d3.v7.min.js D3.js library for data visualization

How It Works

  1. User joins a Zoom meeting with RTMS enabled
  2. Zoom sends RTMS events to the webhook endpoint
  3. RTMSManager processes events and receives transcripts
  4. Each transcript is analyzed by AI to extract personality traits:
    • Curiosity - Questions, exploration, wonder
    • Empathy - Understanding, compassion, support
    • Assertiveness - Confidence, directness, leadership
    • Creativity - Innovation, imagination, originality
    • Analytical - Logic, data-driven, problem-solving
  5. Trait counts are accumulated across the meeting
  6. Results are broadcast to the frontend via WebSocket
  7. D3.js visualization updates in real-time showing trait distribution

Troubleshooting

Issue Solution
No trait data appearing Check OPENROUTER_API_KEY is valid
Visualization not updating Verify WebSocket connection in browser console
Webhook not receiving events Ensure ngrok URL is registered in Zoom Marketplace
JSON parse errors Check AI model is returning valid JSON format

See Also

Docker

The project runs the AI transcript analysis Zoom App. Its multi-stage Dockerfile keeps build tooling out of the final runtime image and does not hard-code a CPU architecture.

Build and run it from the rtms-samples repository root:

docker build -f zoom_apps/ai_transcript_analysis_js/Dockerfile -t rtms-zoom_apps-ai_transcript_analysis_js .
docker run --rm --env-file zoom_apps/ai_transcript_analysis_js/.env -p 3000:3000 rtms-zoom_apps-ai_transcript_analysis_js

Run the build from the repository root because the Dockerfile uses repository-relative paths. Runtime secrets are supplied with --env-file and are excluded from the image build context.

Webhook Delivery Authentication

Normal Zoom webhook deliveries are verified against the exact raw request body using x-zm-signature and x-zm-request-timestamp. Configure ZOOM_SECRET_TOKEN with the Marketplace app's webhook Secret Token. Requests with missing, invalid, or stale signatures are rejected; the default replay window is 300 seconds and can be changed with WEBHOOK_TIMESTAMP_TOLERANCE_SECONDS.