[Suggestion] SvelteKit 2 real-time streaming transcription with server hooks (TypeScript)

[deepgram-robot]

What to build

A working SvelteKit 2 application with real-time streaming transcription using Deepgram's WebSocket API. The app should capture microphone audio in the browser, stream it through a SvelteKit server endpoint (to keep the API key server-side), and display live transcript updates using Svelte 5 reactivity. Include speaker diarization labels and a clean, minimal UI.

Why this matters

SvelteKit is one of the fastest-growing full-stack frameworks with a passionate developer community. Developers building real-time features in SvelteKit need a reference for WebSocket integration patterns, particularly for audio streaming where API key security matters. There's no current SvelteKit example for Deepgram, and the previous Svelte example (circa 2024) predates SvelteKit 2 and Svelte 5's runes-based reactivity model. This fills a visible frontend framework gap.

Suggested scope

• Language: TypeScript
• Framework: SvelteKit 2 with Svelte 5 runes ($state, $derived)
• Deepgram APIs: Streaming STT (Nova-3), speaker diarization
• Architecture: Browser mic capture → SvelteKit server hook (WebSocket proxy) → Deepgram WebSocket
• UI: Minimal Svelte components — mic toggle button, live transcript display with speaker labels, interim/final result differentiation
• Complexity: Intermediate — demonstrates SvelteKit server hooks, WebSocket handling, and Svelte 5 reactivity

Acceptance criteria

• Runnable with minimal setup (clone, add API key, npm run dev)
• README explains the pattern clearly
• Uses SvelteKit 2 and Svelte 5 with runes-based reactivity
• API key stays server-side (never exposed to browser)
• Shows interim and final transcript results
• Includes speaker diarization
• Uses current Deepgram JS SDK version

---

Raised by the DX intelligence system.