MeetMate

MeetMate is a Chrome extension that helps you analyze screenshots, take notes, and summarize your Google Meets effortlessly.

Comment

About MeetMate

Inspiration

As a professional who regularly participates in virtual meetings, I realized the core challenge we all face: the cognitive overload of simultaneously engaging, note-taking, and trying to capture critical visual information. The manual act of switching context to grab a screenshot or furiously type notes invariably leads to missing key action items or losing the thread of the discussion.

MeetMate is the answer to this problem. By leveraging Chrome's Built-in Gemini Nano AI, I provided a professional co-pilot that works entirely on-device. This system handles the heavy lifting, from automatically analyzing visual information in screenshots and summarizing unstructured notes, to offering simplified and translated live captions. This allows the user to be fully present and engaged, without the fear of crucial information being lost.

What it does

MeetMate is an on-device Chrome extension that uses the Built-in AI models to maximize meeting efficiency and accessibility:

  • Smart Notes: Uses the Prompt API to automatically extract action items and summaries from your unstructured notes.
  • Instant Analysis: Uses the Gemini Nano Multimodal API to capture and analyze screenshots, instantly pulling out key visual points and action items.
  • Live Caption Intelligence: Enhances accessibility by using the Rewriter API for simplified captions and the Translator API for real-time translation.
  • Securely stores all notes, analyses, and captions locally on your device for quick review and PDF/Markdown export.
  • Works fully offline for privacy .

Crucially, all processing is 100% private and on-device, leveraging Chrome's Built-in AI without sending any data to a server.

How I built it

I built MeetMate around Chrome's Built-in Gemini Nano, using five key APIs:

  • Prompt API extracts action items from my notes
  • Multimodal Prompt API analyzes screenshots
  • Rewriter API simplifies captions in real time
  • Translator API translates captions live

Everything saves to chrome.storage.local with smart quota management. When storage fills up, the oldest meetings automatically get cleaned up.

Process Flow

MeetMate Demo

Challenges and Breakthroughs

  • Challenge: Multimodal Overload. Processing high-quality screenshots with Gemini Nano risked freezing the browser.
    • Breakthrough: Implemented an efficient multimodal queueing system for background processing, ensuring instant capture and smooth performance.

  • Challenge: Real-Time Reliability. Google Meet's constantly changing structure made reliable caption capture difficult.
    • Breakthrough: Used a MutationObserver for robust caption detection and built a stable asynchronous state management system, guaranteeing real-time accuracy.

  • Challenge: Storage Limits & Data Quality. High-res screenshots quickly consumed local storage, and the Rewriter API needed quality control.
    • Breakthrough: Engineered smart quota management for chrome.storage.local (automatic cleanup) and fine-tuned the Rewriter API prompt to ensure simplification never sacrificed critical meeting information.

Impact and Future

MeetMate demonstrates how on-device AI can deliver sophisticated, privacy-first productivity and accessibility today. Looking forward, my roadmap is focused on scalability, enhanced intelligence, and user choice:

  • Hybrid Mode for Choice: I plan to introduce a user-controlled Hybrid Setting. While the default remains 100% private Gemini Nano, users can opt in to selectively leverage more powerful cloud based LLM models.
  • Intelligent Transcripts: The next major feature is Speaker Identification. Integrating this will transform the current caption archive into truly useful, structured meeting transcripts, labeling who said what.
  • Platform Expansion: I plan to extend MeetMate's functionality to support other major web conferencing platforms like Zoom and Teams, maintaining the same on device as default approach.

Why it matters

This solves something everyone deals with - meeting memory. The difference is we didn't compromise on privacy to do it. Most AI tools send your data away. We proved you don't need to. Gemini Nano running locally is powerful enough for real world use, and it's actually faster because there's no network latency. For users, it's simple: never worry about missing something important again. For the industry, it's proof that on device AI isn't just viable - it's better.

Technical Innovation

  • Smart Quota Management ensures storage stability and private data cleanup on the local device.
  • The Rewriter and Translator APIs process captions as they appear. I built a pipeline fast enough to eliminate all delay, ensuring simplified or translated captions keep pace with natural speech.

Built With

Share this project:

Updates