PodCite

A real-time podcast analysis tool that automatically transcribes audio and extracts notable context, claims, and references as you listen. The application processes podcast content to identify significant statements for multi-source verification. Additionally, users can manually select specific transcript segments to trigger in-depth research workflows.

Demo

Link

Features

• RSS Feed Processing - Parse podcast feeds and download episodes
• Real-time Transcription - OpenAI Whisper-based speech-to-text with timestamps
• Automated Context Extraction - AI automatically identifies notable statements and starts research
• Multi-Source Research - Verifies claims using arXiv, Brave Search, and Congress.gov
• Manual Research - Select any text to trigger custom research

Tech Stack

• Frontend: Next.js, TypeScript, Tailwind CSS
• Backend: FastAPI, Python 3.11+
• AI: LangGraph, xAI API (grok-4-fast), OpenAI (Whisper)
• Search APIs: Brave Search, arXiv, Congress.gov

Prerequisites

• Python 3.11+
• Node.js 18+
• uv - Python package manager (install)
• API Keys (see Configuration below)

Installation

Backend

cd backend
uv sync

Frontend

cd frontend
npm install

Configuration

Create a .env file in the /backend directory:

# Required
OPENAI_API_KEY=...          # Whisper transcription
XAI_API_KEY=...             # Grok-4-fast for AI workflows
BRAVE_API_KEY=...           # Web search
CONGRESS_API_KEY=...        # Congress.gov API

# Optional
LANGSMITH_API_KEY=...       # LangSmith workflow debugging

Running the Application

Start both servers in separate terminals:

Backend (port 8000):

cd backend
uv run main.py

Frontend (port 3000):

cd frontend
npm run dev

Usage

1. Enter RSS Feed URL - Paste any podcast RSS feed URL
2. Episode Loads - Most recent episode loads automatically with audio
3. Start Playback - Click play to begin transcription
4. View Research - Research results appear in the right panel as the podcast plays
5. Manual Research - Select any transcript text to trigger custom research

How It Works

Transcription Pipeline

• Audio is split into 120-second chunks
• First 2 chunks transcribe immediately on play
• Additional chunks load progressively as you listen
• Seeking ahead triggers transcription of that section

Research Workflow

• Every 25 seconds of playback, the system analyzes the previous 20-second window
• AI extracts notable statements (claims, statistics, references)
• Each statement is researched using LangGraph workflows
• Results include source verification, confidence scores, and URLs

Project Structure

├── backend/
│   ├── app/
│   │   ├── api/v1/endpoints/     # API routes
│   │   ├── services/             # Business logic
│   │   ├── workflows/            # LangGraph AI workflows
│   │   └── main.py               # FastAPI app
│   └── media/                    # Audio & transcription cache
├── frontend/
│   ├── app/                      # Next.js pages
│   ├── components/               # React components
│   └── lib/                      # API client