// the find
beltromatti/get-it
Read it. See it. Get it. Built at GDG AI Hack Milan 2026 for "Learn Different" track.
A desktop Electron + Next.js app that takes a PDF or Markdown file and builds a study environment around it: inline concept tags with visualizations, a knowledge graph that scores mastery on four axes, and four study tools (chat, flashcards, quizzes, Feynman). Built in 24 hours at a hackathon and then polished into a shippable product. Aimed at students who already have the source material and want something more than a summary generator.
The bring-your-own-AI model is genuinely clever — routing all LLM calls through the user's own CLI (Codex, Claude, Gemini, or Ollama) sidesteps the 'pay us for tokens' trap that kills every other edtech product. The four-axis mastery scoring (memory, comprehension, structure, application) is a real attempt at measuring depth rather than surface recall, and the monotone non-decreasing clamp prevents score anxiety from sessions where you're tired. The self-repairing sandbox for generated visualizations — where the agent reads its own crash and re-emits a fix — is a practical solution to the flakiness of LLM-generated code. The architecture is honest about what runs where: everything stays local, the only outbound call is an anonymous install ping.
The provider abstraction running everything through CLI subprocesses (Codex CLI, Claude Code, Gemini CLI) rather than HTTP APIs is a fragile dependency chain — any CLI update, auth change, or rate limit surface is outside the app's control, and error handling across subprocess boundaries is always messier than people expect. The 150-page upload cap and batch-of-5-pages concurrency limit means the architecture hasn't been stress-tested against real academic textbooks, which routinely run 400–800 pages. The knowledge graph 'monotone non-decreasing' clamp is a UX choice that sounds reassuring but hides real signal — if you bomb a Feynman session on a concept you thought you knew, the graph won't tell you. Windows builds ship unsigned, which means SmartScreen warnings on every fresh install for a student audience that will absolutely not know to click 'More info → Run anyway'.