AI Video Calling SaaS.
A video calling platform with AI-powered features that turn every meeting into a queryable knowledge artefact — automatic transcription, summarisation, action-item extraction, and per-speaker analytics.
Key features.
- WebRTC-based HD video with adaptive bitrate
- Real-time multilingual transcription with speaker diarisation
- Post-call AI summary with action items, decisions, blockers
- Searchable meeting archive (RAG over past calls)
- Integration hooks for Slack, Notion, Linear (auto-post action items)
Architecture.
Three planes — a media plane handled by a self-hosted LiveKit cluster, a control plane (Node.js API) for room management, and an AI plane (Python) for transcription and summarisation. Recordings persist to S3; transcripts and embeddings to PostgreSQL with pgvector for retrieval.
Tech stack.
- Frontend
- Next.js · WebRTC · Tailwind
- Backend
- Node.js + Python AI service
- Database
- PostgreSQL · S3 for recordings
- AI
- Whisper for transcription · GPT-class summarisation · pgvector for RAG
- Infrastructure
- LiveKit (signaling) · Vercel · AWS for AI workloads
Target users.
Distributed teams, sales orgs, consultants, product managers
Unique selling points.
- Self-hostable — sensitive industries (legal, healthcare) keep data on-prem
- Speaker-diarised transcripts produce accurate per-person analytics
- RAG over your meeting history makes 'what did we decide about X?' instant
Monetization.
Per-seat SaaS + enterprise self-hosting license