AI-Driven Meeting Analysis

Transcript → Summary → Sentiment → Deepfake

Overview

🚧 This project is built for educational purposes by the developer and is still under active development. A lightweight web app that turns recorded meetings into actionable outputs:

• Transcript (Faster-Whisper)
• Summary (BART-SAMSum)
• Sentiment per segment (RoBERTa)
• Deepfake frame scoring (XceptionNet)
• Consolidated PDF export

Frontend: React · Backend: FastAPI · Runtime: CPU-only friendly

Features

• Single-shot processing after upload/record (no streaming)
• Token-aware chunking (~900 tokens) + optional second-pass summary
• Per-segment sentiment labels + aggregate pie chart
• Deepfake scoring for MP4 (≈30s/frame) + frames ZIP export
• Inline status/error banners; no blocking pop-ups

Architecture

Quick Start

Prerequisites

• Python 3.10 (FFmpeg available in PATH)
• Node 18+ / npm
• (Windows) Verified with Intel i5-11400, Chrome

One-click (Windows)

In repo root, double-click: start_fyp.bat

Manual

Backend

cd backend
python -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload