Talixi

Context

Online learning is fragmented — a different course, app, and paywall for every topic, and almost none of it shaped to how you actually learn. Talixi started from one question: what if you could ask for a course on anything and get an expertly structured one in seconds, then have it adapt as you go?

What I built

• →A generation pipeline that turns any topic into a complete course — modules, lessons, quizzes, and a learning path — built on the Vercel AI SDK.
• →Adaptive pathing and an in-lesson AI tutor that adjusts to the learner’s level and answers questions in context.
• →Interactive lessons with syntax-highlighted code (Shiki), math (KaTeX), and embedded quizzes — rendered through my own json-render engine.
• →The full product around it: accounts (better-auth), subscriptions and billing (Polar), and a Postgres data model on Drizzle + Neon.

Highlights

• →A prompt + evaluation pipeline that keeps generated courses structured and accurate across wildly different subjects — from machine learning to music theory.
• →json-render in production: structured AI output rendered into live, interactive Svelte UI instead of plain markdown.
• →Gamification that actually retains — XP, streaks, achievements, and leagues wired into the learning loop.

Outcome

Live at talixi.com — generate a complete, structured course on any topic in seconds, then keep learning with adaptive lessons and an in-context AI tutor.

Links