Overview
Polygo is an AI-powered real-time chat platform that ensures nothing is lost in translation—not even sarcasm, idioms, or cultural nuances. By combining Sarvam AI's Indic translations with Google Gemini's cultural intelligence, it provides an invisible layer of understanding for global communities. Features include multicast translation, sentiment-aware voice (TTS), and real-time AI insights for learning as you chat.
Technical Implementation
Hybrid AI Routing Engine
Developed an intelligent routing system that uses Sarvam AI for blazing-fast Indic translations and Google Gemini Flash for complex global translations and cultural reasoning.
Real-Time WebSocket Sync
Leveraged Convex's real-time backend and WebSocket sync to ensure ACID-compliant, low-latency communication and live-subscription logic for message streams.
Sentiment-Aware Audio Generation
Integrated Sarvam's Bulbul v3 API to generate human-like audio with dynamic pacing that adapts to the emotional sentiment detected in the text.
Pedagogical metadata extraction
Gemini API acts as a language specialist, extracting grammar tips and cultural context as metadata to create a learning-first chatting experience.
Key Features
- Multicast Translation: Single message broadcasted to multiple languages in parallel
- Dual-Core AI Routing: Seamless selection between Sarvam (Indic) and Gemini (Global)
- Cultural Nuance Detection: AI-powered explanations of honorifics and idioms
- Sentiment-Aware Voice: Human-like TTS using Bulbul v3 with emotion-based pacing
- AI Pedagogy: Pedagogical tips from conversations to help users learn while chatting
- Secure Group Auth: Role-based mutation guards in Convex with Auth0 identity
- Admin Controls: Creator-led group management with intuitive Admin labeling
Screenshots
Seamless support for over 22 Indic languages and major global foreign languages.
Multicast group translation showing parallel messages in Hindi, Marathi, Russian, and Tamil.
AI-powered Language Insights explaining cultural nuances and etymology of translated words.
Intuitive chat interface with real-time status updates and simplified group management.
Challenges & Solutions
Challenge: Handling cultural nuances in literal translations
Solution: Implemented a context-aware layer using Gemini Flash to identify idioms and honorifics, providing tooltips that explain 'why' a word was translated in a certain way.
Challenge: Synchronizing translations across many participants simultaneously
Solution: Built a multicast translation pipeline in Convex that triggers parallel AI calls for each member's target language, ensuring near-instant updates for everyone.
Challenge: Ensuring consistent latency for Indic vs Global languages
Solution: Optimized the routing logic to prioritize Sarvam AI's edge-optimized Indic models while using Gemini's global capacity for other languages.
Challenge: Managing authentication across real-time subscriptions
Solution: Implemented role-based mutation guards and secure JWT bridging between Auth0 and Convex to prevent unauthorized message access.