MicDrop AI

• 

  MicDrop Homepage

Inspiration

We were inspired by how powerful podcasts are and how limited they are by time, people, and production. At the same time, AI is usually reactive instead of truly autonomous. We wanted to build a system that could think, research, converse, and perform a podcast by itself.

What it does

MicDrop AI is a fully autonomous podcast generator that creates natural, two-host conversations on any topic using real-world data. It intelligently weaves sponsor mentions into the discussion and continuously improves its own output using memory and self-critique loops.

How we built it

We used Claude for reasoning and dialogue, Lightpanda + Playwright for real-time web scraping, Redis for memory, and ElevenLabs for voice generation. The system runs as a pipeline that researches, writes, improves, stores memory, and speaks the final episode.

Challenges we ran into

Handling JavaScript-heavy sites, maintaining natural conversation flow, and preventing repetitive sponsor mentions were major challenges. We solved these using Lightpanda Cloud, refined prompts, and a Redis-based memory system.

Accomplishments that we're proud of

We successfully built a fully autonomous, self-improving AI podcast system with realistic two-person dialogue. The system researches live data, integrates sponsors naturally, and runs end-to-end with no human scripting.

What we learned

Autonomy requires memory, feedback, and strong system design. We learned how to turn LLMs into agents with identity, improve outputs through self-critique, and engineer AI behavior instead of just calling an API.

What's next for MicDrop AI

We plan to add auto-publishing to Spotify, video/avatar podcasts, multi-language support, and live episode scheduling. Our goal is to evolve MicDrop AI into a full autonomous media platform.

Built With

• anthropic
• flask
• lightpanda
• react
• redis
• sanity