$adarsh-jha-dev~projects/podcastr
back to projects
live·2024

~/podcastr

AI-powered podcast generation platform

Next.jsTypeScriptConvexOpenAITailwindClerk

// overview

Generate full podcasts from a transcript — pick an AI voice, generate or upload a thumbnail, publish. Mobile-first, real-time backend on Convex.

// problem

Creating a podcast is a multi-tool dance: scripting, recording, editing, designing cover art, hosting. I wanted to compress that into a single flow where the only input is the script.

// approach

Built on Next.js App Router with Convex as the realtime backend — it gave me reactive queries out of the box. Wired OpenAI's TTS for voices and an image model for thumbnails. Clerk handled auth so I could focus on the content pipeline.

// highlights

  • AI voice synthesis from text transcripts
  • AI thumbnail generation with image models
  • Real-time data sync with Convex DB
  • Fully responsive across mobile & desktop

// learnings

Convex is genuinely magical for projects that need realtime without managing websockets. But streaming AI output through serverless functions has cold-start tax — worth measuring before committing.