AutoPod

Podcast Prompt
Podcast Library

Inspiration

AutoPod started from a simple observation: people absorb information better through audio, but creating a podcast is still much harder than consuming one. Even short, educational episodes require structured research, clear writing, consistent tone, multiple voices, and audio production skills. We wanted a way for anyone, students, educators, or independent creators, to take a topic they want to learn or explain and instantly turn it into a clean, engaging podcast without needing technical knowledge. The inspiration came from seeing how much time we and others spend studying or reviewing content, and imagining how helpful it would be if an AI system could do the heavy lifting and deliver it in a format you can listen to anywhere.

What it does

AutoPod turns any topic into a fully produced, multi-speaker podcast episode. Users enter a topic, select a tone and duration, and AutoPod handles the entire pipeline automatically. Claude validates and cleans the topic, generates detailed research notes, and writes a full script for three speakers with names, personalities, and emotional context on every line. The TTS system converts each line into speech using Cartesia’s Sonic-3 voices, then merges the audio into a seamless MP3 episode stored in Supabase. Users can log in through Supabase Auth, save episodes, replay them in their dashboard, and generate as many podcasts as they want.

How we built it

We built a modular backend with Express, TypeScript, and Claude Sonnet-4.5 for topic validation, research generation, and script writing. Structured outputs ensure Claude returns clean JSON that the backend can convert directly into audio instructions. The TTS service uses Cartesia’s API, generating speech for each script line with emotion tags, then merging everything into one final audio file. Supabase handles user authentication and episode storage. The frontend is built with Next.js, and Tailwind, giving users a chat-style interface for generating episodes and a clean dashboard to access their saved content.

Challenges we ran into

The hardest part was designing prompts that produce stable, structured scripts with consistent speaker names, pacing, and emotions. Multi-speaker formatting broke easily without very strict instructions. Another challenge was aligning the research output with the script generator so that Claude stayed factual and didn’t drift. TTS required careful handling because each line needed to match the correct voice and emotion, and merging the audio segments into a smooth final track took time to get right. Integrating Supabase storage and making sure files were publicly accessible was also more tedious than expected.

Accomplishments that we're proud of

We’re proud that AutoPod produces podcast episodes that actually feel natural and coherent, not stitched-together AI text. Getting three speakers, emotional context, and tone-specific delivery working end-to-end was a major achievement. We also built a clean, reliable pipeline from prompt → research → script → TTS → storage, and it all works with a simple user interface. Seeing a full MP3 episode generate from scratch in one request feels like the exact kind of tool we wanted when we started the project.

What we learned

We learned how valuable structured outputs are when building reasoning systems. Without strict schemas, multi-speaker content falls apart quickly. We also learned how important prompt design is when generating narratives that need to sound natural when spoken aloud. On the technical side, we became comfortable integrating LLMs with real workflows, building TTS pipelines, merging audio, and handling Supabase auth and storage. The project taught us how to design infrastructure that keeps Claude’s reasoning predictable and clean.

What's next for AutoPod

Next, we want to support more voices, more languages, and optional sound design like intro music or ambient effects. We also want to add a feature that turns full documents or webpages into podcast episodes, since many people want audio summaries of long materials. Longer-term, we want to build a collaborative mode where two users can create a podcast together and Claude acts as the “third speaker,” helping structure the conversation. AutoPod can extend far beyond education, and we want to keep expanding it into a general tool for audio-based reasoning and knowledge sharing.