WiserPin - Pin and summarize web content with AI
Inspiration
We all save bookmarks with good intentions, but they quickly become digital graveyards—hundreds of links we'll never revisit because we can't remember what they contained. Traditional bookmarks only save URLs and titles, not the actual knowledge inside. We wanted to build a smarter way to save the web that respects user privacy while delivering intelligent features.
What it does
WiserPin is a privacy-first bookmark manager that automatically summarizes web pages as you save them. It's a browser extension and web dashboard that:
- Instantly summarizes any webpage using AI when you bookmark it
- Organizes intelligently by auto-suggesting the right collection based on content
- Keeps everything private by processing summaries locally with Google Chrome's built-in Gemini Nano
- Works offline-first with local-model processing, falling back to cloud AI only when needed
- Makes rediscovery easy with searchable summaries and organized collections
Instead of saving just a link, you capture the key insights from every page—and your data never leaves your device.
How we built it
Tech Stack:
- Browser Extension: TypeScript, Chrome Extension Manifest V3, Plasmo framework
- Web Dashboard: React, TypeScript, Vite, TailwindCSS, shadcn/ui components
- Backend API: NestJS, TypeScript, Prisma ORM
- Database: PostgreSQL on Google Cloud SQL
- Infrastructure: Google Cloud Platform (Cloud Run, Cloud Storage, Load Balancer, Terraform)
- AI: Gemini Nano (on-device via Chrome AI APIs) with Gemini 1.5 Flash fallback
Architecture:
- Local-model first approach: Gemini Nano runs in the browser for privacy
- Cloud fallback: Gemini 1.5 Flash API for browsers without Nano support
- Extension captures pages and generates summaries client-side when possible
- API handles authentication (Clerk), data persistence, and cloud AI fallback
- Static web dashboard served from Cloud Storage with backend via Cloud Run
Challenges we ran into
Chrome AI API Limitations: Gemini Nano's availability is limited to specific Chrome versions and requires experimental flags. We had to build a robust fallback system to detect availability and gracefully switch between local and cloud AI.
CORS and Extension Security: Getting the extension to communicate securely with our API while maintaining Chrome's strict Content Security Policy was tricky. We had to carefully configure permissions and implement proper authentication flows.
Prompt Engineering: Getting consistent, high-quality summaries from both Gemini Nano (local) and Gemini Flash (cloud) required extensive prompt tuning to ensure similar output quality across both models.
Infrastructure Complexity: Setting up a production-grade GCP deployment with SSL, load balancing, Cloud Run, and Cloud Storage while keeping costs low for a hackathon project required careful architecture decisions.
Collection Auto-Suggest: Building an AI feature that analyzes content and suggests the best collection required understanding the nuances of few-shot prompting and structured output formatting.
Accomplishments that we're proud of
True Privacy-First AI: We successfully implemented on-device AI summarization using Gemini Nano, proving that powerful AI features don't require sacrificing user privacy.
Seamless Fallback System: Built an intelligent detection system that uses local AI when available and seamlessly falls back to cloud processing, providing a consistent experience across all Chrome versions.
Production-Ready Infrastructure: Deployed a complete full-stack application on GCP with proper SSL, load balancing, database, and authentication—all documented and reproducible via Terraform.
Polished User Experience: Created both a functional browser extension and a beautiful web dashboard that makes bookmark management actually enjoyable.
Real-World Utility: Built something we'd actually use daily—WiserPin solves a genuine problem with bookmarks becoming unusable over time.
What we learned
Chrome AI APIs are the Future: Working with Gemini Nano showed us the potential of on-device AI. The ability to run LLMs locally in the browser opens up entirely new categories of privacy-respecting applications.
Local-First Architecture: Designing for local-first processing while maintaining cloud fallbacks taught us valuable lessons about progressive enhancement and graceful degradation.
Prompt Engineering Matters: Small changes in prompts dramatically affect output quality. We learned the importance of clear instructions, examples, and constraints when working with LLMs.
Extension Development Evolution: Chrome's Manifest V3 represents a significant shift in extension architecture. Understanding service workers vs. background pages was crucial.
Infrastructure as Code Pays Off: Using Terraform from the start made our GCP deployment reproducible and debuggable, saving hours of manual configuration.
What's next for WiserPin - Pin and summarize web content with AI
Enhanced Search: Implement semantic search across summaries using embeddings, allowing users to find bookmarks by concept rather than just keywords.
Smart Collections: Auto-create collections based on browsing patterns and content themes without user intervention.
Collaboration Features: Allow teams to share curated collections with private summaries, creating collaborative knowledge bases.
Multi-Browser Support: Expand beyond Chrome to Firefox and Edge, adapting the local AI strategy for each platform's capabilities.
Mobile Apps: Build native iOS and Android apps that sync with the browser extension for on-the-go access.
Advanced AI Features:
- Auto-tagging based on content analysis
- Related bookmark suggestions
- Automatic duplicate detection
- Reading time estimates and priority scoring
Export and Integration: Enable export to note-taking apps (Notion, Obsidian, Roam) and integrate with read-it-later services.
Analytics Dashboard: Show users insights about their saved content—topics they're researching, reading patterns, and knowledge gaps.
Log in or sign up for Devpost to join the conversation.