[Suggestion] Vercel Edge Function streaming audio transcription (TypeScript)

[deepgram-robot]

What to build

A Vercel Edge Function that accepts audio data and returns transcription results from Deepgram's STT API. Demonstrates how to run Deepgram transcription at the edge with minimal cold-start latency, suitable for serverless voice applications.

Why this matters

Edge computing is increasingly important for voice applications where latency matters. Developers deploying on Vercel need a reference implementation showing how to handle audio processing in Edge Functions, which run closer to users with sub-100ms cold starts. This pattern is valuable for real-time transcription APIs, webhook-based audio processing, and serverless voice application backends. Vercel is one of the most popular deployment platforms for modern web apps, and an edge transcription example makes Deepgram accessible to this large developer community.

Suggested scope

• Language/Framework: TypeScript / Vercel Edge Functions (Next.js App Router)
• Deepgram APIs: Pre-recorded STT (Nova-3) via REST API
• Features: Edge Function receives audio → forwards to Deepgram → returns transcript. Includes a simple frontend page for testing with file upload and microphone recording.
• Complexity: Low-medium — Edge Function + simple React frontend
• Deployment: Vercel (free tier compatible)

Acceptance criteria

• Runnable with minimal setup (clone, add API key, vercel dev)
• README explains the edge architecture pattern clearly
• Uses current Deepgram JS SDK or direct REST API compatible with Edge Runtime
• Edge Function runs in Vercel Edge Runtime (not Node.js runtime)
• Includes simple frontend for testing audio upload
• Demonstrates response handling from Deepgram back to client

---

Raised by the DX intelligence system.