manabit.ai

A full-stack Japanese learning platform I built solo and use every day while prepping for the JLPT. It pairs AI-generated learning content with a spaced-repetition engine and per-user progress tracking.

Most language-learning apps are either generic — the same vocabulary decks for everyone — or rigid, with review schedules that don't care how well you actually remember something. I was studying for JLPT N2 and wanted content that stayed relevant and a review schedule that adapted to how I was actually doing.

manabit.ai generates vocabulary and reading content with AI, then uses a spaced-repetition algorithm driven by each learner's own performance to schedule review, so material comes back right before you're about to forget it.

Next.js App Router for both the front end and API routes, PostgreSQL for durable content and review-history storage, with an AI provider integration for content generation. Deployed on Vercel.

Keeping AI-generated content good without letting generation cost and latency get out of hand, and tuning the spaced-repetition schedule so review load stays manageable without hurting retention.

Using the tool myself every day turned out to be the fastest feedback loop I could ask for. Friction I'd never notice in a demo is impossible to ignore after a week of actual use.