Hi HN. I built Lex in an attempt to overcome the same hurdle that I always faced as a language learner: lots of vocabulary drilling, and no place to read or write. I'd find myself having to use a flashcard app, reading on my own, and then trying to find a space to write and none of them could communicate with each other. Lex combines the three to get 34 languages:

Reading: leveled texts (CEFR A1-C2) or AI generated texts on a topic of your choice. Tap on any word when you read it to add it directly to your vocabulary! Writing: generate and/or compose an essay on a custom essay prompt, and receive AI feedback on structure, errors and an estimated essay level. Vocab: words that you save are used for spaced-repetition exercises (Flashcards, Match pairs, Fill-in Blanks, etc.).

For the generation and essay feedback, I used OpenAI, and for front and back I used Next.js + Postgres. But it's early and there is much I have yet to do. Here are two things I'm totally unsure about and would love to hear HN's thoughts on:

Quality of the AI essay feedback depends a lot on language, tuning prompts per language is harder than I thought. Unit economics: each essay check/retext creation is an API call and people want a language app to be cheap. Not quite sure of the sustainable pricing.

I am polyglot and speak a few languages myself, so this is very much a tool that I created out of my own frustration. This is as much feedback as they can get, and every "this already exists, it's called X" is welcome.