Nov 21, 2026

Tatva: Meetings End, the work doesn't

murf.ai agora tatva voiceai ai

Teams make decisions in meetings and then lose them: notes are incomplete, nobody remembers who owns what, and follow-ups are never sent. Tatva closes that loop.

Tatva is a voice-first agentic meeting assistant. A Chrome extension records browser meetings, and Tatva turns each one into a searchable team knowledge base: summary, decisions, action items with owners and due dates, numbers mentioned, key moments and follow-up meetings, plus any PDFs or slides shared in the meeting.

Users then talk to Pooja, Tatva's voice agent built on Agora Conversational AI and Murf AI Falcon. Pooja answers questions grounded in that knowledge and cites her sources, then gets the follow-up work done across Gmail, Google Calendar, Slack and Google Sheets. Every critical action is read back and needs the user's approval, errors are explained, and every action is logged.

How it works: the Manifest V3 extension captures tab audio and microphone through tabCapture and an offscreen document, and uploads a WebM recording to our Node.js and TypeScript server. Groq Whisper transcribes it with timestamps, and a Groq LLM extracts strict JSON (decisions, owners, dates, figures, follow-ups). Transcripts, extractions and attachments (PDFs parsed, images read by a vision model) are chunked, embedded locally with transformers.js and stored in Supabase pgvector for retrieval. For voice, Agora Conversational AI runs the real-time loop over Agora RTC: voice activity detection, turn-taking, barge-in and Deepgram speech-to-text in Indian English. Through Agora's custom LLM integration, every turn reaches our OpenAI-compatible streaming endpoint, which performs RAG and tool calling on Groq; actions run through Composio after a deterministic confirmation step. Replies stream back to Agora and are spoken by Murf AI Falcon TTS (voice Pooja, Indian English, conversational style, 24 kHz), giving natural, low-latency speech. Typed questions use Agora's messaging channel and the same agent pipeline.

Our goal: no decision lost after a meeting, and measurable time saved for every team, shown per meeting.

This build was uploaded as a hackathon project

Hackathon

Build with Agora - BLR

View All Projects
Give a star to encourage!Discussion
Start a new conversation!
Login to join the discussion