overview.
Vartaa is an AI-driven team communication platform that enables instant translation across 22 Indian languages. It combines Sarvam AI’s linguistic models with Convex’s real-time backend for zero-friction collaboration.
technical implementation.
Hybrid AI Routing
Optimized latency by routing Indic translations to Sarvam AI (Mayura:v1) and complex global reasoning/summarization to Google Gemini Flash.
Real-Time Sync Engine
Leveraged Convex’s WebSocket architecture for ACID-compliant, <100ms synchronization of messages, typing indicators, and live translations.
Sentiment-Aware Audio
Integrated Sarvam Bulbul v3 for TTS, using a custom analysis layer to dynamically adjust voice pacing and tone based on message sentiment.
Automated Metadata Extraction
Deployed Gemini-powered insights to automatically extract actionable tasks, smart replies, and professional tone alerts from unstructured chat.
key features.
- Real-time Translation: Automatically translate messages into every member's preferred language
- AI Thread Summaries: Catch up on long conversations with Gemini-powered summaries
- Universal Voice Messaging: Automatic transcription, translation, and playback in chosen languages
- AI-Extracted Tasks: Automatically identify and extract action items from chat messages
- Ultra-Low Latency Smart Replies: Context-aware suggestions with under 200ms response time
- Focused Rooms: Zero-noise architecture that eliminates notification fatigue
- Real-time Infrastructure: Live typing indicators, presence tracking, and read-receipts
screenshots.
High-performance team collaboration with AI-driven summaries
Focused rooms designed for zero-noise professional communication
Multilingual messaging with <100ms synchronization across devices
Instant native translation across 22+ Indian languages
challenges & solutions.
Challenge: Handling many translations without slowing down the app
Solution: Built a system that processes multiple languages at the same time in the background, so the chat stays fast and smooth for everyone.
Challenge: Choosing between fast translations and smart summaries
Solution: Used two different AIs: one that is super fast for basic translating and another that is very smart for creating summaries and finding tasks.
Challenge: Showing real-time "typing" updates without any lag
Solution: Used high-speed technology to sync updates across all users in less than a second, so everyone sees instantly when someone is typing.
Challenge: Making the complex design look premium on small mobile screens
Solution: Created a custom layout that automatically reshapes itself for phones, keeping the "Obsidian" look beautiful on any device.
Challenge: Remembering important work mentioned in a long chat
Solution: Used AI to automatically find and save "Tasks" and "To-Do" items from the messages, so nothing important gets forgotten.