🧠✨ NeuroCanvas

NeuroCanvas is a full-stack interactive neural network project that lets users draw digits on a canvas, train a simple neural network in real-time, and predict handwritten digits — all within the browser using a Python backend.

It's an end-to-end demo of how neural networks work behind the scenes — from pixel input to prediction — without using any machine learning libraries like TensorFlow or PyTorch.

---

🎯 Features

• ✍️ Canvas-based Drawing: Draw digits directly in the browser on a 20x20 grid
• 🧠 Custom Neural Network: Pure Python feedforward neural net implementation (no ML libraries!)
• 🔁 Train & Predict: Train the model in real-time via frontend and instantly test predictions
• 🛠 Full-Stack Setup:
  • Frontend: HTML5 + JavaScript Canvas
  • Backend: Python http.server with custom NN logic
• 💾 Model Persistence: Save and load weights to/from JSON
• 🧪 Based on MNIST (Optional): Easily switch to training on MNIST data

---

🚀 Demo

• checkout the ss or run it.

---

📁 Folder Structure

1. ocr.py - has the offline training from the MNIST set
2. neural_network_design.py - neural network training (also test and predict)
3. server.py - (Backend) receives response after sending to the nn
4. ocr.html
5. ocr.js
6. Other loaded and saved files of the model and the backpropagated results fro optimised use