Meetly
overview.
Meetly is an encrypted video conferencing platform featuring host lobbies, client-side E2EE streams, and an automated post-meeting AI pipeline that generates summaries and transcripts using Sarvam AI and Grok-3.
technical implementation.
Serverless Post-Meeting Pipeline
Configured Upstash QStash to securely trigger an offloaded background worker, orchestrating sequential API calls to Sarvam and xAI without blocking user flows.
Client-Side Cryptographic Workers
Implemented client-side encryption using LiveKit's worker scripts to encrypt WebRTC media tracks in transit based on local cryptographic hash keys.
Indic-Optimized Audio Transcription
Integrated Sarvam AI's speech-to-text API (optimized for Indian English) to convert multi-party meeting recordings into high-fidelity transcripts.
Dynamic Quality Adaptability
Built custom hooks monitoring system-constrained client events to automatically lower publisher/subscriber stream qualities and preserve performance.
key features.
- End-to-End Encryption (E2EE): Encrypted media channels utilizing client-side workers and URL hash passphrases.
- Host-Controlled Lobby: Waiting room security checking access tokens and prompting host approval.
- AI Speech-to-Text: Native integration with Sarvam AI optimized for Indian-accented English audio.
- Grok-3 Meeting Summaries: Automated generation of structured executive summaries, decisions, and next steps.
- QStash Background Queue: Resilient asynchronous job processing for meeting recording analysis.
- Real-time Moderation: Full panel interface for muting individuals, muting all, or removing participants.
- Adaptive Performance Engine: Low-CPU mode that automatically downscales streaming quality under system constraints.
screenshots.
Secure lobby gatekeeping allowing hosts to verify and admit participants before entering
High-performance WebRTC meeting layout with End-to-End Encryption enabled via URL key exchanges
Real-time participant management panel offering muting, kicking, and room-wide moderation
Post-call summary page displaying Grok-generated decisions, action items, and collapsible transcripts
challenges & solutions.
Challenge: Managing high-latency AI transcription and summarization calls on API routes
Solution: Decoupled the pipeline using Upstash QStash queues to execute background jobs asynchronously and handle auto-retries on transient failure states.
Challenge: Ensuring AI summaries strictly conform to application data schemas
Solution: Programmed strict system prompts for Grok-3, enforcing JSON-only structures containing summaries, key decisions, action items, and next steps.
Challenge: Synchronizing participant entry state across hosts and clients
Solution: Established a lobby verification endpoint that issues short-lived participant tokens only after host admission is registered.
Challenge: Avoiding rendering lag during multi-user screen and video share operations
Solution: Utilized adaptive streaming, dynacast, and a single peer connection configuration in the LiveKit client configuration.