📻 RepoRadio

AI-Generated Podcast Reviews of GitHub Repositories

Turn any GitHub repository into an entertaining, educational podcast with multiple host personalities, background music, sponsor breaks, and professional audio production.

---

🎯 What is RepoRadio?

RepoRadio analyzes GitHub repositories and generates natural-sounding podcast conversations between tech experts discussing the code, architecture, dependencies, and interesting technical details. Think "tech podcast meets code review."

Key Features

🎙️ 8 Unique Host Personalities

• Alex (Hype Developer), Casey (Tech Enthusiast), Jordan (Pragmatic PM)
• Morgan (Cynical Senior), Quinn (Security Paranoid), Riley (Junior Optimistic)
• Sam (DevOps Wizard), Taylor (Open Source Purist)

🎬 Production Studio Features (All enabled by default)

• 🎵 Background lo-fi music with auto-ducking
• 🎺 Intro/outro jingles
• 📢 Humorous sponsor ads based on project dependencies
• 🎚️ Crossfade transitions between speakers
• 🔊 Professional audio mixing

🤖 Intelligent Analysis

• Git archaeology (recent commits, top contributors)
• Dependency scanning (package.json, requirements.txt, etc.)
• Deep mode: AI-selected priority files for code review
• Daytona sandbox for safe repo cloning

🔊 Voice Options

• Local: Kokoro ONNX (fast, private, free)
• Cloud: ElevenLabs (premium quality)

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• Ollama with llama3.1:8b model
• Daytona for sandboxed repo access

Installation

1. Clone and install ```bash git clone cd hacka python -m venv venv source venv/bin/activate # Windows: venv\Scripts\activate pip install -r requirements.txt ```

2. Set up Ollama ```bash ollama pull llama3.1:8b ```

3. Configure environment ```bash cp .env.example .env

Edit .env and set your DAYTONA_API_KEY

```

4. Run ```bash ./start.sh # or: streamlit run src/app.py ```

---

🎮 Usage

1. Open http://localhost:8501
2. Paste a GitHub URL
3. Select 1-3 hosts from 8 personalities
4. Click GENERATE VIBE
5. Listen to your podcast! 🎧

---

🏗️ Architecture

Script Generation (One-Line-at-a-Time)

• Generate ONE line at a time (8-12 iterations)
• Pre-break: 4-6 lines introducing the project
• Ad break: 1 sponsor announcement (if enabled)
• Post-break: 4-6 lines diving into technical details
• 100% reliable - no more truncation issues

Audio Production Pipeline

1. Parallel TTS (4 workers, ~10s for 12 lines)
2. Crossfade transitions (200ms)
3. Sponsor ad transition sounds
4. Background music overlay (-20dB)
5. Intro/outro jingles
6. Export to MP3

---

🧪 Testing

```bash PYTHONPATH=src:$PYTHONPATH pytest tests/ -v ```

All 15 unit tests passing ✅

---

📝 Documentation

• PRODUCTION_STUDIO.md: Audio features guide
• DEBUG_LOGGING.md: Logging system
• src/music/README.md: Music folder organization

---

🛣️ Roadmap

• Cloud LLM support (OpenAI/Anthropic)
• Export to YouTube/SoundCloud
• Multi-language support
• Voice cloning for custom characters

---

🤝 Contributing

Contributions welcome! Areas: characters, ad templates, music, tests, performance.

---

📄 License

MIT License

---

🙏 Credits

• Kokoro TTS: hexgrad/Kokoro-82M
• Daytona: Sandboxed dev environments
• Ollama: Local LLM inference
• Streamlit: Web UI

---

RepoRadio: Where code meets conversation 🎙️✨