InkSolver
The open-source AI whiteboard that solves STEM problems as you draw them. Sketch an equation, circuit, or reaction on an infinite canvas and watch verified, step-by-step solutions appear.
Overview
InkSolver turns a blank canvas into a tutor. Draw a problem in handwriting, shapes, text, or a pasted image, and a vision model reads it and streams a structured, step-by-step solution back in real time. It is built for the way people actually think through STEM: messy sketches first, clean answers after.
The frontend is Next.js 15, React 19, and TypeScript on a tldraw infinite canvas, with KaTeX rendering the math. The part I'm proudest of is the trust layer: a FastAPI + SymPy backend symbolically verifies every solution step, so the AI's reasoning is checked rather than taken on faith. Solutions export to PDF, PNG, or LaTeX and share via read-only links, with semantic search over past work powered by Postgres + pgvector.
Architecturally it is a single Next.js deployment orchestrating vision inference and symbolic verification as swappable service modules, so it degrades gracefully when a service isn't configured. The project is currently in active development, with a public deployment on the way.