SiteSage

Inspiration

Democratizing AI access is a big task and perhaps the most important one if we want to have AI enhance all our lives. What we find today is that implementing AI is costly and not affordable or easy for small businesses. We aim to change that.

What it does

This solution allows anyone with a website to host a customized AI chatbot that handles user queries.

How we built it

We use the power of RAG (Retrieval Augmented Generation) and Gemini Nano built right into the browser, thanks to Google. All of this runs locally on the user's brower. So no need to pay for any API usage or fine tune a model. Webmasters can create the context file easily and have a customized chatbot right in their website that provides answers custom tailored to their business.

Challenges we ran into

Getting proper context out of a large text file is hard. We used fuzzy searching, but it can be improved with more techniques like synonym finding, advanced fuzzy searching, piping the match score to the LLM, etc.

Accomplishments that we're proud of

We are able to get decent context out of the file without using any external API calls. The speed and accuracy of the generation looks good.

What we learned

AI in Chrome is super powerful and giving the APIs out means that developers are going to create custom uses which are useful to people. Also using the API is very easy so that a web developer with a basic knowledge of AI can easily pick up.

What's next for SiteSage

We want to make the context matching better with using synonmys and advanced fuzzy searching. The chatbot will be made such that context and history is maintained over page navigation.

Built With

• chrome
• fuse.js
• gemini
• javascript