KhmerMeet AI
KhmerMeet AI is a production-minded MVP for Cambodian teams to record meetings, paste or generate transcripts, create Khmer/English summaries, extract action tasks, assign owners, set deadlines, and track progress.
Features
- Email/password auth with protected dashboard
- Browser audio recording with local file storage
- Browser WebRTC video meetings for local MVP testing
- OpenRouter transcription for Khmer and English meeting audio
- OpenRouter meeting summaries in Khmer by default
- OpenRouter JSON action task extraction
- Meeting history, meeting detail, task management, settings
- Responsive SaaS dashboard UI with Khmer-friendly typography
- Text export for transcript and summary
Tech Stack
- Next.js App Router, TypeScript, Tailwind CSS
- Prisma ORM with
DATABASE_URL
- NextAuth credentials auth
- OpenRouter API
- Local file storage for MVP
Setup
- Install dependencies:
npm install
- Copy environment variables:
cp .env.example .env
- Fill in:
DATABASE_URL="postgresql://postgres:postgres@localhost:5432/khmermeet_ai?schema=public"
OPEN_ROUTER_API_KEY="..."
OPEN_ROUTER_TEXT_MODEL=openai/gpt-4o-mini
OPEN_ROUTER_TRANSCRIBE_MODEL=google/chirp-3
OPEN_ROUTER_TRANSCRIBE_FALLBACK_MODEL=google/gemini-2.5-pro
OPEN_ROUTER_TRANSCRIBE_TIMEOUT_MS=55000
NEXTAUTH_SECRET="replace-with-a-long-random-secret"
NEXTAUTH_URL="https://khmermeet-ai.vercel.app"
- Create the database and seed demo data:
npm run prisma:migrate
npm run prisma:seed
- Start local dev:
npm run dev
Demo login after seeding:
Production Deployment
Before giving KhmerMeet AI to real users, follow the deployment checklist in DEPLOYMENT.md. It lists the required Vercel environment variables, PostgreSQL setup, Supabase Storage bucket, OpenRouter checks, and LiveKit configuration.
Notes
The MVP stores audio files under uploads/ locally and serves them through /api/uploads/[name]. On Vercel, local uploads are temporary, so production should move this adapter in lib/storage.ts to S3 or Supabase Storage.
Server Rec records each participant's microphone locally in their own browser (a room-wide signal over the LiveKit data channel starts it automatically for everyone once one person clicks the button) and posts segments straight to the transcription API - no S3-compatible storage is required and no extra environment variables need to be set. This replaced an earlier LiveKit Egress + S3 upload design after hitting a confirmed, unresolved AWS SDK V2 signature-validation bug in Supabase Storage's S3-compatible API (see https://github.com/supabase/storage/issues/646); lib/livekit-egress.ts and the /api/livekit-egress/* routes are kept in the codebase unused in case that upstream bug gets fixed and server-side recording becomes worth revisiting.
Video meetings run entirely on LiveKit Cloud (WebRTC SFU) - it handles STUN/TURN relay for reliable calls across different networks internally, so no separate TURN server configuration is needed.
OpenRouter powers transcription, summary generation, and task extraction. OpenRouter is not a media server; video/audio transport is handled by LiveKit.
Future Improvements
- Real-time speech-to-text
- Speaker detection
- Google Calendar integration
- Telegram reminders
- Slack and Zoom integrations
- PDF export
- Team workspace
- Payment plan
- Cloud storage
