SnapCal (Offline AI Study Assistant)

AI-powered math and study assistant that works completely offline using Chrome’s built-in on-device AI APIs. Learn, solve, and understand even without internet.

Comment

Inspiration

In Pakistan, internet instability, load-shedding, and protests often shut down access to online tools. As a student, I’ve personally faced situations where the internet went out right before an exam or assignment. My younger brother also struggled to study math without online help. These experiences made me realize how dependent we’ve become on connectivity and how unfair that is for students in low-connectivity areas. That’s what inspired SnapCal, a tool that brings AI-powered learning completely offline, using Chrome’s on-device Gemini Nano model.

What It Does

SnapCal is an AI-powered study assistant that helps users solve math problems, explain concepts, and understand study material even without the internet. It works in two modes:

  1. Web App: Type or upload questions (text or image) and get instant explanations.
  2. Chrome Extension: Right-click any question on a webpage → “Solve with SnapCal” → get an AI-powered answer instantly.

All inference happens locally using the Prompt API and on-device Gemini Nano, so it’s private, fast, and works 100% offline.

How We Built It

  • Frontend: Pure HTML, CSS, and JS (no heavy frameworks).
  • AI Core: Chrome’s experimental window.ai.languageModel.create() and Gemini Nano.
  • Extension: Uses a context menu and background scripts to send selected text to the AI model.
  • Multimodal Support: Text + image questions processed locally.
  • Setup: Works on Windows (tested on AMD RX 580 GPU) with Chrome Dev build and on-device AI flags enabled.

Challenges We Ran Into

  • Getting Gemini Nano to install and load properly the documentation was minimal, and enabling the right Chrome flags was tricky.
  • Debugging constant ai is not defined and permission errors in the extension.
  • Balancing web app and extension development with API stability.
  • Handling multimodal input and on-device inference without cloud fallback.
  • Limited access to real-time resources due to Pakistan’s frequent connectivity issues — ironically, the exact problem SnapCal solves.

Accomplishments That We’re Proud Of

  • Built a fully offline AI study assistant that works without internet.
  • Successfully integrated Chrome’s on-device AI APIs before they’re widely documented.
  • Made it run smoothly on mid-range consumer hardware (no dedicated AI chip).
  • Demonstrated that AI accessibility is possible for developing countries without cloud dependence.

What We Learned

  • How to use and debug Chrome’s on-device AI stack and Gemini Nano models.
  • How to design for low-resource environments while keeping UX smooth.
  • The importance of local inference for privacy, reliability, and education equality.
  • Gained deeper understanding of multimodal input handling and context window optimization.

What’s Next for SnapCal (Offline AI Study Assistant)

  • Add voice input using Chrome’s On-Device Speech Recognition API.
  • Expand to multilingual support for Urdu and regional languages.
  • Release a lightweight mobile version with on-device Gemini Nano Lite.
  • Open-source the project fully with setup instructions for other low-connectivity regions.
  • Collaborate with educational NGOs to make offline AI learning tools widely accessible across Pakistan and beyond.

Built With

Share this project:

Updates