Inspiration
SearchMate was inspired by our own struggles with technology and inconveniences we have experienced firsthand as consumers. As students with hundreds of files on our computers, it's near impossible to find my notes on one specific topic from two weeks ago without manually opening and scanning multiple files in my explorer. Studying for a test involves finding relevant notes across multiple documents and copy-pasting them into a study guide, and finding that one screenshot I saved as the default "IMG_808720DI" is a nightmare.
What it does
To save users time and increase convenience, we developed a desktop extension that runs locally as a smart-search assistant. Users can make a query for information they want from their computer without needing to know the name of the file(s), such as "find me my biology notes from a few weeks ago on mitosis cell division". It also works with images, so users can find "photo of me and my family in front of statue in Italy" without having to scour their folders and click through each photo to identify it. SearchMate can also consolidate information on the same topic across files and browser history to save you the time it takes to do it manually- so you can automatically create those study guides straight from your notes, find the most relevant photos for your infographic, or get a high-level summary of a project from multiple meeting notes.
How we built it
We built the backend using Python, specifically working with CLIPModel for embeddings. The frontend was built using JavaScript, utilizing the Electron framework to bring a native experience to users.
Challenges we ran into
One of the challenges we faced was how to integrate the UI in a way that felt natural in the context of the OS, given our background in web-development. To address this, we brought in the Electron framework, which allowed us to tailor the user interface more intuitively. Another challenge we faced was figuring out how to deploy the embeddings model within the Intel framework. We went through many trials and referenced example code to learn how to do it and apply the knowledge to our use case.
Accomplishments that we're proud of
We were able to integrate a fully working embedding model that runs locally, as well as devise a dynamic indexing model to search for files.
What we learned
Throughout the development process, we learned how to be resourceful in utilizing open source code within our own frameworks and adapting usage of external models to our needs. We also learned the importance of communication and effective teamwork.
What's next for SearchMate
We hope to see SearchMate in the hands of users across Intel's customer base, changing the way people study, work, and access files. Your information, your way.
Built With
- clipmodel
- electron
- flask
- javascript
- pydentic
- python
Log in or sign up for Devpost to join the conversation.