Inspiration
- Growing popularity of podcasts as a medium for content consumption.
- Aiming to blend the best of written and audio content into personalized podcasts.
What it does
PodcastGPT allows users to create personalized podcasts. Features include:
- Selection of a celebrity host.
- Customizable podcast duration and genre.
- Option to input personal news articles or use an in-built news search.
- Podcasts mimic host mannerisms.
- Ability to upload podcasts to YouTube.
How we built it
- OpenAI Assistants API for podcast agent configuration.
- LlamaIndex and Tavily Search for news aggregation.
- Custom YouTube integration for podcast uploads.
- ElevenLabs for celebrity voice recreation.
Challenges we ran into
- Managing the context window for large article inputs.
- Ensuring deterministic news retrieval via Tavily Search.
Accomplishments that we're proud of
- Integrated OpenAI API, LlamaIndex, Tavily Search, ElevenLabs, and YouTube, achieving seamless functionality.
- Mastered ElevenLabs' voice recreation, adding unique celebrity voices to podcasts.
- Developed a method for efficient article condensation, addressing AI context window limits.
- Enabled reliable and relevant news retrieval through Tavily Search.
What we learned
- Learned to navigate and mitigate agent limitations, particularly in content processing.
- Understood the subtleties of voice technology for creating engaging audio content.
- Recognized the importance of user-driven customization in content platforms.
What's next for PodcastGPT
- Introduction of customized voices and self-created hosts.
- Integration with Spotify for podcast sharing.
- Expanding content ingestion capabilities (e.g., YouTube videos, PDFs).
Built With
- llamaindex
- openai
- python
- tavily

Log in or sign up for Devpost to join the conversation.