Inspiration
Our inspiration comes from AI's ability to explain concepts and suggest new content to explore. We hope our extension can not only inspire people on where to visit next but also keep the web alive in a time when AI is becoming an aggregator and web managers are losing traffic volume at a rapid rate.
What it does
As you browse the web, Gemini Rec Search scans your content and leverages Gemini's advanced capabilities to generate precise Google search queries. Instantly, you're presented with additional webpages tailored to you—allowing you to find content across the web that helps you.
Features
Smart Content Discovery and Content Summary
- Seamlessly discover relevant content as you browse
- Get intelligent recommendations based on your current page context
- Save time finding related resources across the web
- Get content summaries to enhance understanding
Summary Glance
- Get a summary of a website before visiting
- Simple and intuitive by clicking the Star button
Powered by Gemini AI
- Advanced content analysis for precise recommendations
- Real-time query generation tailored to your interests
- High-quality results powered by Google's latest AI technology
Simple & Privacy-Focused
- No buttons to press, works automatically when you open.
- Full control over the domains the extension can access
How we built it
We built Gemini Rec Search utilizing ReactJS with Typescript and Gemini Nano on-device model. Our extension runs entirely on-device utilizing a Chrome Canary build and the Gemini Nano model, but it also can run on-server using a Gemini API call. The Chrome Extension simply asks for permission to read the webpage content, and for each new domain you visit, we ask for permission to read from that domain. This gives the user the ability to restrict our extension to certain domains.
The extension saves user data on-device, utilizing Chrome storage to save allowed domains. Thus, the user's web searches and visits are private and only accessible to them.
Challenges we ran into
We ran into a few challenges; our biggest challenge was an error that Gemini Nano would throw about the output language not being English. We decreased the number of times we encountered the error by stripping our input of non-English characters. Another challenge we ran into was Gemini Nano only being available under the early preview and only on Chrome Canary; this limited our testing environment and ensured a similar environment for all our users.
Accomplishments that we're proud of
One of the key achievements we resonate with the most is the usability of the extension. This extension solves a problem that we both faced and we have been key users of our extension for the past couple weeks. We are happy that we got to build something that we use ourselves and can introduce to our friends and families
What we learned
The development process allowed us to learn quite a lot - especially about on-device models and even more about the Gemini family of models. This experience allowed us to understand more about the LLMs that have taken over the world for the recent past couple years.
We've become proficient in Chrome Extension creation and prompting AI models to get the results we would like. This experience allows us to build even more extensions in the future.
What's next for Gemini Recommended Search
We are hoping to launch the Chrome Extension using a server for people who do not have access to Gemini Nano and then transition to Gemini Nano as it becomes more available. We also hope to increase the accuracy and dependability of our news bias rating.
Log in or sign up for Devpost to join the conversation.