Problem
Large images take up meaningful storage and bandwidth. Could unsupervised learning compress an image while keeping it usable, and could that be turned into something people can actually use rather than a notebook nobody opens?
Approach
Used K-means clustering to group similar pixel colors together, reducing the number of distinct colors needed to represent an image and shrinking its storage footprint. Rather than stopping at a notebook, the model was deployed as a Streamlit web app so users can upload their own pictures, compress them, and download the result.
Result
A working, publicly deployed compression tool. Not just a proof of concept but a shipped app people can use today, currently supporting BMP images.