What I Learned Building EchoNote

I started EchoNote because I had a problem myself. It's the best reason to build something - and the only one that holds when motivation disappears.

#The idea came from frustration

I was listening to 5-10 podcasts a week. Common thread: when I heard something useful, I took manual notes. It worked terribly. I missed half the points while writing. What I did capture was fragmented. And none of it was ever retrieved.

#Tech stack: what I chose and why

Next.js (App Router) for frontend and API routes. One codebase, rapid iteration.

Supabase for database, auth, and storage. PostgreSQL with Row Level Security meant I didn't think about database security from day one.

OpenAI Whisper for transcription. Open-source alternatives struggled with Danish. OpenAI's hosted version was faster and more accurate.

GPT-4 for summarization and chapter division. The most expensive part, but also the highest value.

#What I didn't foresee

Danish is hard for AI. Whisper quality on Danish varies wildly depending on audio quality.

People want control, not automation. My first prototype was fully automatic. Users hated it. They wanted to edit, annotate, and customize. Version 2 made everything interactive.

Pricing is the hardest part. I settled on simple: free up to 3 episodes/month, then flat monthly. Simple enough to move on and focus on the product.

The things I thought mattered, didn't. I spent two weeks on advanced chapter detection. Users were indifferent. They wanted: fast transcription, accurate summaries, and search across old episodes.

#Where EchoNote is heading

Building EchoNote into a knowledge platform. Listen to podcasts, EchoNote builds your personal knowledge base. "What have I heard about AI regulation in the last 6 months?" - that question EchoNote should answer.

---

Try EchoNote - upload your first podcast and see what AI finds.