Audorapod is an AI-powered podcast generation tool that transforms any topic into professional podcast scripts and audio content. Using parallel research via Exa.AI and Azure OpenAI, it creates natural conversational scripts with 3-host formats and optional voice synthesis via ElevenLabs.

• Fast Research: 8 parallel searches via Exa.AI with AI synthesis
• Natural Scripts: Conversational podcast scripts with 3-host format
• Voice Generation: Text-to-speech with ElevenLabs (optional)
• Flexible Duration: 1-60 minute episodes that scale automatically
• Easy Customization: All AI prompts in /prompts/ folder for easy editing
• No Confirmation: Runs directly without prompts

1. Install Dependencies

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install -r requirements.txt

2. Environment Setup

   • Copy .env.example to .env and fill in your API keys:

     cp .env.example .env

   • Required: Exa.AI API key, Azure OpenAI credentials
   • Optional: ElevenLabs API key (for audio generation)

cd src
python main.py --topic "Your topic here"

# 5-minute podcast (default)
python main.py --topic "AI safety trends"

# 10-minute podcast
python main.py --topic "Future of remote work" --duration 10

# With audio generation
python main.py --topic "Climate tech" --duration 15 --audio

• --topic or -t: Podcast topic (required)
• --duration or -d: Length in minutes (1-60, default: 5)
• --audio or -a: Generate audio with ElevenLabs TTS
• --verbose or -v: Detailed logging
• --cleanup or -c: Remove audio segments after merging

The application generates files in the output/ directory:

• research_<topic>_<timestamp>.json - Research data from Exa.AI
• script_<topic>_<timestamp>.json - Podcast script in JSON format
• script_<topic>_<timestamp>.txt - Readable podcast script
• audio/ - Individual MP3 segments (when using --audio)
• podcast_full_<timestamp>.mp3 - Complete merged podcast

Topic + Duration → Research → Synthesis → Script Generation → Audio (Optional)
       ↓              ↓           ↓             ↓                  ↓
   User Input     Exa.AI API   Key Themes   Natural Dialogue   ElevenLabs TTS
                  (32 sources   Analysis     (Scaled to time)   (3 voices)
                   parallel)

audorapod/
├── venv/                    # Virtual environment
├── src/
│   ├── main.py             # CLI application entry point
│   ├── config.py           # Configuration and API keys
│   ├── azure_llm.py        # Azure OpenAI client (increased token limits)
│   ├── researcher.py       # Exa.AI research module
│   ├── research_synthesizer.py # Research synthesis for script focus
│   ├── scriptwriter.py     # Podcast script generator (natural conversations)
│   ├── voice_generator.py  # ElevenLabs TTS integration
│   ├── audio_merger.py     # Audio file merging with FFmpeg
│   └── prompt_loader.py    # Utility for loading external prompts
├── prompts/
│   ├── researcher_prompt.md # Research analysis prompt
│   ├── scriptwriter.md      # Script generation prompt
│   ├── query_generator.md   # Search query generation prompt
│   ├── synthesis.md         # Research synthesis prompt
│   └── system_prompt.md     # System prompt for JSON responses
├── output/                  # Generated podcast files
└── .env                     # API configuration

1. User Input: python main.py --topic "AI safety trends" --duration 10
2. Research Phase: 8 parallel searches via Exa.AI (~32 high-quality sources)
3. Synthesis Phase: AI extracts key themes and conversation hooks
4. Script Generation: Natural conversational script with 3-host format
5. Audio Generation (optional): Text-to-speech conversion using ElevenLabs
6. Output: Files saved to output/ directory with timestamps

Typical execution time: ~1.5 minutes for complete workflow (80% faster!)

All AI prompts are in the /prompts/ folder - edit them to customize behavior:

• query_generator.md - How search queries are created
• synthesis.md - How research is synthesized into themes
• scriptwriter.md - How conversations are generated
• researcher_prompt.md - How research is analyzed
• system_prompt.md - Basic system instructions

• Fast: 8 parallel searches, ~1.5 minute execution time
• Hosts: Sarah (host), Tony (expert), Brian (trend analyst)
• Natural: Authentic conversations with interruptions and reactions
• Scalable: Episodes automatically adjust to requested duration
• Simple: No confirmation prompts, runs directly

• FFmpeg: Required for audio merging (install via Homebrew: brew install ffmpeg)
• ElevenLabs Account: For voice generation (configure voice IDs in .env)

If you encounter import or dependency issues:

1. Ensure you're using the correct Python environment: source venv/bin/activate
2. Verify OpenAI version: pip list | grep openai (should be 1.35.0)
3. Try alternative execution method: PYTHONPATH=src venv/bin/python -m main