TabTab

Superhuman for Tabs: Browser Assistant that evolves with you!

Comment

Inspiration

As digital-first beings and enthusiasts passionate about learning and research across various topics, we end with too many open tabs on browser every day. This is not only killing the browser performance, but our own productivity and performance. As we generate these digital footprints leaving petabytes of data across the web, we are unable to gain insights into our own web journey and use it to enhance our potential.

What it does

Presenting to you TabTab: A browser assistant aka chrome extension that empowers you, by effectively organizing your tabs and providing valuable insights into your browsing and learning journeys and how they evolves over the time by:

  1. Providing a chronological timeline of the activity
  2. Dynamically categorizing the topics explored
  3. Generating summaries - providing a quick recap of the journey
  4. Intelligently archiving the tabs based on category and priority for seamless revisits

How we built it

We started with locking the MVP features and building mocks in Figma. For the product development, we used MongoDB Atlas, LlamaIndex, and Together.AI to semantically understand the visited website's content with embeddings. We started categorizing the visited websites leveraging Claude. We used built in chrome features to setup the TabTab extension.

Challenges we ran into

  1. [Eng Challenge] Finding the optimal solution for categorization of websites' content
  2. [Design Challenge] We strived to perform effective categorization: distinct and succinct website links under each category. The tradeoff we considered is querying the web content embeddings Vs clustering the content embeddings. The query approach would not guarantee that links within each category are mutually exclusive while the clustering approach is not widely supported
  3. [Eng Challenge] Claude access issues through web extension for summarization
  4. [Design Challenge] A lot of websites require logged-in/paid access to understand the content. Our current access limits us with only publicly available sites.

Accomplishments that we're proud of

  1. We created vector embeddings of our web traversals leveraging MongoDB Atlas and LlamaIndex, as part of building RAG
  2. We were able to get together as a team instantly and locked in the product design and roadmap within a short period time. Throughout, we succeeded as a team in supporting one another, learning from one another in building TabTab
  3. We have a functioning chrome extension demonstrating end-to-end flow of our TabTab

What we learned

  1. Crawling the website is much more challenging than personal data/files, since a lot of websites have dynamically generated content as well as block/rate limit bot traffic including server side generated content that we can't retrieve. Thus, we understood why rewind.ai used the accessibility API's in real time
  2. Better evaluation of models

What's next for TabTab

In an ideal world, we would love to gain customer feedback on our MVP so far to make further product decisions. Just prioritizing our current research so far we want to:

  1. Chat interface to provide deeper and quicker insights into web journey
  2. Dynamically organize the tabs across windows for easier navigation across also reducing duplication
  3. Switch to Claude for efficient and high performant ways to generate summaries directly from the web URLs
  4. Finish smart archiving feature of the tabs

Built With

Share this project:

Updates