Lumi — Spoken English Practice

Lumi is a warm, encouraging spoken-English practice companion built on AssemblyAI. Talk for a minute, review the words and patterns you used, then practice out loud with an AI tutor.

Live demo

Try it on Vercel: https://assembly-ai-demo.vercel.app

Allow microphone access when prompted so Lumi can capture live speech.

What you can do

1. Warm up — Pick a topic (weekend, coffee order, work, or free talk) and see your streak and collected words.
2. Speak — Lumi listens live via AssemblyAI Streaming STT and shows gentle captions as you talk.
3. Takeaways — Get vocabulary, sentence patterns, and a practice prompt generated from your transcript.
4. Practice — Talk with the AssemblyAI voice agent using your lesson materials.
5. Wrap up — Celebrate your streak and words saved for the day.

Progress (streak, words collected, sessions) is stored in your browser via localStorage.

Setup

Create .env.local:

ASSEMBLYAI_API_KEY=your_key_here

Install and run:

bun install
bun run dev

Open http://localhost:3000.

Test flow

1. Choose a topic and tap Start speaking.
2. Speak English for 10–20 seconds and confirm captions appear.
3. Tap I am done speaking, then See my takeaways.
4. Review vocabulary and patterns, then Practice with Lumi.
5. Tap Start practice with Lumi, have a short exchange, then Finish session.
6. Confirm streak and word counts update on the wrap-up screen.

For engineering details, open Developer diagnostics at the bottom of the app.

Product plan

See Product Plan.md for UX flow, design language, and the reusable UI-generation meta-prompt.

Stack

• TanStack Start + SolidJS
• Tailwind CSS v4
• AssemblyAI Streaming STT (universal-streaming-english)
• AssemblyAI Voice Agent
• Local study-material generation (/api/study-materials)