Inspiration

I noticed that most people - especially students - spend countless hours reading articles, research papers, and study materials online, yet struggle to retain what they learn. Learning on the web is largely passive - you read, scroll, and forget. I wanted to change that.

Meet Knowva - a new way of learning on web - one where reading on web becomes interactive and memorable.

What it does

Knowva is a Chrome extension that helps readers turn web content into interactive flashcards and quizzes in a few seconds. By transforming passive reading into active recall, it enables students to remember more without leaving their browser.

You can highlight any text or right-click an image, and Knowva’s Gemini Nano-powered multimodal AI will automatically generate learning materials such as:

  • Flashcards focusing key points
  • Quiz to test your understanding
  • Instant feedback based on your performance

Everything runs locally on your device, ensuring complete privacy and seamless performance.

How I built it

I discovered the problem of passive reading and I wanted to build an experience that was different from existing Large Language Model (LLM) - user interacts directly inside the browser instead of copying the content, switching tabs and pasting content to LLM.

I built Knowva using:

  • Frontend: Chrome Extension with side panel integration
  • UI and state management: React + TypeScript + TailwindCSS + ShadcnUI
  • Gemini Nano (via Chrome Prompt API) for on-device, multimodal AI processing

The workflow:

  1. Capture selected text or image metadata through Chrome’s context menu API.
  2. The data is sent to a local AI session using the Prompt API.
  3. The AI generates structured flashcards and quizzes, which render in the extension’s side panel using React components.
  4. Users can interact, test, and get instant feedback - all offline & right in the browser.

Challenges I ran into

  • Limited documentation for Gemini Nano: Working with an experimental on-device model required a lot of trial and error.
  • Fine-tuning model: I had to experiment with different system prompt to produce more accurate result.

Accomplishments that I'm proud of

  • Successfully integrated Gemini Nano for offline, multimodal AI processing within a Chrome extension.
  • Built a fully privacy-preserving learning tool - no cloud, no tracking, all local.
  • Instead of building another LLM, I created an experience that guides the user to turn webpage content into a personalized study session in seconds.

What I learned

  • How to build a web extension from scratch with React
  • How to work with emerging Chrome APIs like the Prompt API and on-device model support.
  • How to make a difference with existing technologies such as ChatGPT (Why user should use this instead of ChatGPT)

What's next for Knowva - Turn Passive Learning into Active Learning

  • Support audio and video using multimodal capabilities such as Podcast and YouTube video.
  • Support for additional languages
  • More customization options:
    • To export flashcards
    • Difficulty level for quiz (easy, medium, hard)
    • AI-generated feedback based on performance (user knows what to improve)

Built With

Share this project:

Updates

Submission history