PodSong Mini

PodSong Mini is a cutting-edge AI Audio Studio platform capable of automatically turning any topic in the form of text/url/document (PDF) into a two-way podcast scenario.

Comment

PodSong Mini

Inspiration

Have you ever stared at a dense, ten-page PDF document or a highly complex online article and wished you could just "listen" to it instead? Better yet, what if that dry information could be transformed into a fun, interactive, two-way podcast, wrapping up with a catchy, customized song summarizing the main points?

Departing from the struggle of consuming heavy, text-based content, PodSong Mini was born. It is designed to turn the tedious chore of reading into an engaging, multi-sensory, and highly entertaining auditory experience.

What it does

PodSong Mini is a cutting-edge AI Audio Studio platform that automatically transforms any text, URL, or PDF document into an interactive, two-way podcast, complete with a customized musical summary.

Key Features:

  • 🎙️ Dynamic AI Hosts (Zephyr & Charon): Far beyond standard Text-to-Speech, the platform dynamically scripts real-time debates and discussions between Zephyr (the charismatic host) and Charon (the analytical expert).
  • 🎵 Song Synthesis: A unique innovation that takes the core concepts of your topic and transforms them into poetic lyrics and an epic music masterpiece to close out the episode.
  • 📄 Smart PDF & Context Extraction: Instantly parses and understands the context of complex documents and web links to use as the foundation for the podcast.
  • 🌐 Polyglot Engine: Supports multi-language generation with natural, native accents, eliminating the rigid and robotic tone common in older TTS models.

How we built it

To bring this vision to life, we combined cutting-edge AI models, robust backend logic, and a unique visual identity:

  • The Frontend & Design: Wrapped in a stunning, high-contrast Avant-Garde Brutalism user interface.
  • Context Extraction: Powered by Gemini 2.5, which acts as the brains to read, digest, and generate natural, conversational podcast scripts.
  • The Audio Pipeline (Hybrid TTS Engine): Built with an advanced fallback auto-rotation architecture.
    • Primary Engine: Utilizes Gemini 3.1 Flash TTS for ultra-expressive, native-sounding voice synthesis.
    • Smart Rotation: Implemented intelligent API key rotation to handle request thresholds.
    • Fallback Logic: Connects seamlessly to Google Cloud Text-to-Speech as an automatic backup.

Challenges we ran into

The biggest hurdle we faced was behind-the-scenes API limitations. Because Gemini 3.1 Flash TTS is a cutting-edge technology still under review and subject to strict request restrictions, we frequently ran into quota limits.

Designing a system that could handle these interruptions without breaking the user experience was incredibly tough. We had to engineer a highly resilient, zero-downtime hybrid architecture that detects quota exhaustion instantly and shifts to Google Cloud TTS mid-stream without the user ever noticing a pause.

Accomplishments that we're proud of

  • Zero-Downtime Hybrid Architecture: Successfully building a self-healing audio generation pipeline that auto-rotates API keys and gracefully falls back to secondary services under heavy loads.
  • Authentic AI Chemistry: Getting Zephyr and Charon to sound like real, contrasting individuals engaged in a genuine, intellectual debate rather than two robotic voices reading lines.
  • The "Song Synthesis" Feature: Successfully bridging the gap between text comprehension and musical generation to create catchy, poetic audio summaries.

What we learned

  • Resilient AI Engineering: We learned how crucial it is to design "graceful degradation" into AI applications. When relying on early-access or preview APIs, you must build robust fallbacks to keep the app production-ready.
  • Multi-Agent Prompting: We gained deep insights into structuring multi-turn conversations between different AI personas to ensure natural flow, interruptions, and chemistry.

What's next for PodSong Mini

  • 🎸 Expanded Musical Genres: Allowing users to choose the genre of their summary song (e.g., Synthwave, Lofi, Rock, or Acoustic).
  • 👥 Custom Host Creation: Giving users the ability to design their own hosts, adjusting their tone, expertise level, and personality traits.
  • 📈 Production Scaling: Transitioning the voice synthesis entirely to high-throughput production models as Gemini 3.1 Flash TTS moves out of preview limits.

Built With

Share this project:

Updates