Margin: The Lazy Way to Solve Math with Open-Source AI

Doing math on a computer has always felt like unnecessary hard work. If you want to solve an equation, you usually have to fight with equation editors, type out awkward LaTeX syntax, or jump across three different tabs between a graphing calculator, a textbook, and a scratchpad. Even with pen and paper, the moment you make a small mistake on line two of a long derivation, you end up wasting twenty minutes doing tedious algebra before realizing your answer is completely off.

I built Margin because I wanted to be lazy about math.

The complete source code for Margin, including the canvas engine, the SymPy verification service, and the vision pipeline, is available on GitHub: https://github.com/lochanharishwar/MYPREP

Margin is an infinite digital whiteboard where you do not have to type anything. You just pick up a stylus or use your mouse, scribble down a problem naturally, and let open-source AI and symbolic computing do the rest. As soon as you pause, Margin reads your handwriting, verifies every algebraic step with a deterministic math engine, plots interactive coordinate graphs, and speaks back to you with a patient, friendly conversational tutor that explains where you went wrong. You do not even have to read a giant wall of text because the tutor talks through the lesson out loud while writing the verified steps right onto the board for you.

You can try out the live web application here: https://gen-lang-client-0249652862.web.app/notebook/default

You can also run the entire application locally on your machine with the Next.js frontend and the local Python FastAPI microservice.

Margin is built around a clean separation between generative open-weight vision and deterministic symbolic math:

1. The Vector Canvas: Built with Next.js 14, React, and an HTML5 Canvas that captures handwritten coordinates and smooths raw points into anti-aliased quadratic curves in real time.

2. Open-Weight Vision and Handwriting OCR: When you pause your writing, the canvas isolates the bounding box of your strokes and converts it into a lightweight, high-contrast image.

This gets processed by Google Gemma 4 and fast multimodal models configured specifically for character disambiguation, distinguishing things like the variable x from multiplication, or the letter t from a plus sign.

3. Deterministic Verification with SymPy: Generative models are great at reading pictures, but they are notoriously unreliable at arithmetic.

To keep the math strictly correct, Margin sends every recognized equation to a local Python FastAPI service running SymPy.

SymPy evaluates algebraic equivalence, computes exact derivatives, finds roots, and confirms the derivation step by step.

4. Interactive Display and Graphs: The frontend uses KaTeX to render clean math cards and MathJS to render interactive Cartesian coordinate graphs beside your handwritten work.

5. Conversational Voice Tutor: Mathematical explanations are phonetized into natural conversational English and vocalized using streaming neural speech synthesis, giving you a real-time teaching assistant that speaks like a patient instructor rather than a robotic text reader. Why Does Open Innovation Matter?

6. Open innovation was not just an option for this project, it was the only way to make it work reliably.

First, closed commercial LLMs try to do everything inside a single black box. When you ask a closed model to solve a multi-step calculus problem, it often hallucinates plausible-looking numbers that are mathematically wrong. Open-source innovation allows us to combine the best open-weight vision models with SymPy, a mature, deterministic computer algebra system that has been refined by the global scientific community for over a decade. That pairing gives you actual mathematical proof instead of a statistical guess.

Second, open innovation protects privacy and keeps costs at zero. When students and engineers work on personal scratchpads, their handwritten notes and ideas should not be fed into a proprietary training pipeline or locked behind a twenty-dollar monthly subscription. Because Margin is built on open standards, lightweight SQLite, and modular Python services, anyone can clone the repo, run it on their own laptop, and have a permanent personal math tutor without relying on closed corporate servers.

This project was built and debugged collaboratively using an autonomous coding agent session. The agent assisted in scaffolding the Next.js 14 App Router, writing the vector stroke rasterizer, fine-tuning the SymPy verification microservice, and solving audio race conditions with AbortControllers so speech never overlaps when drawing quickly.

The lazy payoff is huge. Instead of spending ten minutes formatting an integral in a document editor or getting stuck on line four of a quadratic formula because of a missed negative sign, you just write the problem down and sit back. Margin instantly tells you if your intermediate step is algebraically equivalent to your previous one, graphs the curve automatically so you do not have to plot points by hand, and reads the answer out loud. It completely eliminates the tedious busywork of math, letting you move slow, stay relaxed, and focus purely on understanding the concept.

Prize Categories -Overall Open-Source AI Challenge -Best Local or Offline-First AI Application -Best Educational or Productivity Tool