Mohammad Haroon Portfolio · Case study
Meeting AI
AI-Powered Audio Intelligence Platform
Transforms audio recordings into accurate transcripts, intelligent summaries, and searchable conversations.
Overview
Meeting AI is a production-grade full-stack platform that automates the end-to-end meeting workflow for remote and hybrid teams. It combines live WebRTC conferencing, audio recording, OpenAI Whisper transcription with English translation, GPT-powered notes, automated task assignment, and a contextual AI chatbot behind organization-based access control.
About the project
A full-stack AI platform that turns live meetings, uploaded recordings, and Google Meet captures into searchable transcripts, structured notes, assigned tasks, and grounded chatbot answers. Built for teams that need clarity from every conversation without manual note-taking.
Role & contribution
As Full Stack Developer, Mohammad Haroon built the React frontend, Node.js/Express APIs, WebRTC signaling, Chrome extension capture flow, Supabase multi-tenancy, and the Whisper/GPT processing pipeline.
Frontend & backend
Frontend technologies
Backend technologies
Challenges & solutions
WebRTC peer connections
ProblemReliable peer-to-peer audio across NATs and multi-participant rooms is fragile under real network conditions.
SolutionBuilt room-based Socket.io signaling with SDP/ICE routing, RNNoise noise suppression, and a Web Audio mixer for host-side recording.
Chrome MV3 recording limits
ProblemManifest V3 service workers cannot capture Google Meet tab audio directly.
SolutionOrchestrated tab capture via getMediaStreamId and an offscreen RecordRTC document that uploads the recording for processing.
Multi-language meetings
ProblemMixed-language conversations need accurate English transcripts teams can actually search.
SolutionUsed Whisper translations for English output, then GPT-4o mini sentence correction for clarity without losing meaning.
Grounded meeting chat
ProblemGeneric LLM answers can invent details that never happened in the meeting.
SolutionGround every reply in transcript, notes, tasks, and prior chat history before calling the model.
Project details
- Live WebRTC rooms with Socket.io signaling
- Whisper transcription and GPT meeting notes
- Automated task extraction for participants
- Contextual AI chatbot grounded in meeting data
- Chrome extension for Google Meet capture
- Organization-scoped collaboration with Supabase RLS
Full technology stack
Frequently asked questions
What is Meeting AI by Mohammad Haroon?
Meeting AI is a full-stack WebRTC and AI meeting platform that turns live calls and recordings into transcripts, GPT notes, tasks, and searchable conversation context.
Which technologies power Meeting AI?
The stack includes React, Vite, Node.js, Express, Socket.io, Supabase, OpenAI Whisper, GPT-4o mini, and a Chrome MV3 extension for Google Meet capture.
How does Meeting AI keep chatbot answers accurate?
Chat replies are grounded in the meeting transcript, generated notes, extracted tasks, and prior chat history before the model responds.