Filedex

picture of the site indexing a folder
searching for a picture of the moon

Inspiration

Google search is an everyday commodity. Simple, fast and intuitive knowledge just from a search. Why shouldn't your own computer be just as accessible? Using AI embedding models, Filedex sorts your files and allows you to look them up just like a web search, while maintaining full privacy due to its offline nature. Quick and comfortable, Filedex is the file explorer upgrade you never knew you needed.

What it does

Filedex is a seamless integration to upgrade your file explorer experience. Using text and image embedding models and generative AI, it transforms regular search into something akin to a google search for your files - enabling easy lookup and making file lookup easier than ever.

How we built it

FileX is a modular semantic file search system with a Python/FastAPI backend, a React/TypeScript dashboard, and a Git-like local metadata store in a .filex/ directory. The backend generates embeddings for files, stores them locally, and serves fast similarity search over an in-memory index.

Semantic embeddings: Text is embedded with Sentence-Transformers for semantic search workflows, then retrieved via similarity matching. Images are embedded using OpenAI’s CLIP ViT-B/32 model to match images and text in the same embedding space.

FastAPI exposes endpoints for indexing/search and can run heavier work as background tasks so the UI stays responsive. Embeddings are saved on disk (NumPy arrays) and loaded into an in-memory cosine-similarity search index for near-instant queries after indexing.

Challenges we ran into

We have had issues with finding our best embedding models to use on files. It took a lot of tries to figure out what is the most all-encompassing embedding method.

Accomplishments that we're proud of

To have created a fully functional product that significantly improves user experience for a essential part of their personal computer usage.

What we learned

We learned how to use embedding models to normalise text and images so they can both be searched for in the same way, making finding certain files, regardless of file type, much simpler and more intuitive. Additionally, using figma we were able to learn how to make a solid interactive gui that is representative of what we want to let the users experience.