amm_real

Fellow language tool builder here.

1. I signed up and generated a B2 essay in French. Quality was totally fine, but I have to ask: if you're having trouble tuning prompts for the 30+ languages you've enabled, why are you starting with so many languages? Focus down until you hit traction with one or a handful that you yourself know well.

2. I wouldn't worry about unit economics based on the type of LLM use I noticed in the app.

More broadly: From going through my reading exercise, I got the sense actually that you're much more focused on writing and feedback on writing quality. Is that true?

I'm not aware of anything that outright replaces the Anki-style DB/SRS with an LLM or agentic workflow, and I'm not sure that would be a good idea, either. Even study/knowledge apps built from the ground up during the LLM era (like RemNote (https://www.remnote.com/) use an Anki-style SRS AFAIK. But maybe RemNote is what you're after.

Since you're asking about LLMs, there are plenty of Anki extensions that make use of the tech for card creation and supplementation but leave the core database and scheduling loop untouched. Anki Brain (https://ankiweb.net/shared/info/1915225457) comes to mind. My own extension for studying Chinese, Pindu (https://readpindu.com/), uses the output from Anki's SRS as the input to the LLM, and then the learner sees the output from the LLM.

Nice write-up. It would be cool to see codex side-by-side as well. (Maybe you'd have to use a driver other than Haiku.)