MNIST Digit Classifier — AI Web App

Interactive digit recognition with 98% accuracy using PyTorch and Streamlit

Project Overview

Built an interactive digit recognition tool using a custom PyTorch model with 98% accuracy and real-time predictions in Streamlit. Designed a clean, responsive UI with dual-input, confidence scores, probability visualizations, and model explainability. Successfully deployed on Streamlit Cloud for public access.

Technologies Used

Python

PyTorch

Streamlit

Streamlit

Matplotlib

Matplotlib

Key Features

Custom PyTorch neural network achieving 98% accuracy on MNIST dataset
Real-time digit prediction with interactive drawing canvas
Dual input methods: canvas drawing and image upload
Confidence scores and probability distribution visualization
Model explainability with prediction confidence metrics
Clean, responsive UI design optimized for user experience
Cloud deployment on Streamlit Cloud for public access

Challenges & Solutions

Model Optimization

Achieved 98% accuracy through careful architecture design and hyperparameter tuning while maintaining fast inference times.

Real-time Visualization

Implemented efficient probability visualization using matplotlib and streamlit components for smooth user interaction.

Cloud Deployment

Successfully deployed the application on Streamlit Cloud, ensuring reliable public access and performance optimization.