I'm Ali, and I built a tool to help me learn Spanish. There's a good amount of info in the about page, but the short version is that I found apps like Duolingo completely useless. In theory they're designed to help you speak a language but I found it mostly helped my muscle memory tap the right parts of a screen without really learning to say anything. Or more importantly prepare me for how fast a native Spanish speaker speaks!

After giving up on Duolingo I tried a few different things; Pimsleur, Paul Noble and Language Transfer. These are all great in different (but similar) ways and after listening to them in the car for a few months, the next time I went to Spain I was able to hold (very) basic conversations and even book a table at a nearby restaurant over the phone without saying a single word of English.

I also tried ISSEN (a YC startup I saw posting here that advertises itself as an AI language tutor) and found it to be... underwhelming. It spoke at much too high a level for me and I think asked me about pets or something? When I said I didn't have a dog it completely ignored that and asked what breed of dog it was. I've used a few different "voice-native" AI agents and even the ones in English left a lot to be desired in terms of interruptions, understanding and memory. It also doesn't help with the one time in my life that I actually have time to learn Spanish, which is when I'm driving my car.

So I decided to try and recreate the Language Transfer/Paul Noble style learning approach, but letting you write an AI prompt to create a lesson suited to your interests and learning level.

https://cadencespanish.com/

Any feedback would be great!

In terms of tech stack, I used Lovable and Supabase. I wrote some of the copy by hand but none of the code. I use ElevenLabs for the realtime speech-to-text and Google Cloud for the text-to-speech (it's an awful lot cheaper and sounds quite good). I am a software developer (I work for a totally unrelated YC backed startup) but I've somewhat fallen in love with Lovable for how fast I can go from "huh, I wonder if this would work" to something in production. And ironically I learned the most about React and react-router-dom from a previous time when Lovable built something that went so wrong I had to dig into the code and rewrite bits of it.