Cache-22

Cache 22
Calendar Event Detection Feature

Inspiration

When we arrived at this venue, we assumed there would be internet. That turned out to be a bad assumption. It got us thinking: how often do you find yourself wanting to go online, only to realize you can’t? On a long flight? Deep in the subway? At a hackathon with overloaded Wi-Fi? Or halfway through a road trip in the middle of nowhere? That’s why we built Cache-22 — a smart download extension that predicts which websites you’ll want to visit and saves them ahead of time, so your web is there when the web isn’t.

What it does

Cache-22 analyzes your browser history and currently open tabs using AI-powered clustering to understand your browsing patterns and interests. It then intelligently caches related web pages (HTML/CSS + JavaScript) so you can access and interact with them while offline.

How we built it

We built this as a Chrome extension using Manifest V3, leveraging Chrome's storage APIs and content scripts for page extraction. The core intelligence comes from Claude AI, which analyzes browsing patterns and clusters URLs into meaningful categories.

Smart Search Integration: We integrated Bright Data's search API to intelligently discover related content. When Claude identifies your browsing patterns, it generates targeted search queries that are sent to Bright Data, which returns relevant URLs from across the web. This allows us to find and cache content you haven't even discovered yet.

Dual Caching Strategy: For maximum reliability, we implemented a hybrid caching approach using both Playwright (for complex JavaScript-heavy sites) and simple HTML scraping (for faster, lighter pages). This ensures we can capture everything from interactive web apps to static documentation.

Intelligent Auto-Caching: After the initial analysis, the extension automatically triggers whenever you visit new pages, using Claude to analyze the content and generate fresh search queries. This creates a self-improving system that gets smarter about your interests over time.

Offline-First Architecture: All cached content is stored locally using Chrome's storage APIs, ensuring complete offline functionality without relying on external servers or internet connectivity.

Challenges we ran into

One of the toughest challenges was getting the caching system to handle a wide range of websites — from static documentation to dynamic, JavaScript-heavy apps. Ensuring Playwright captured everything accurately without breaking site functionality took a lot of fine-tuning.

We also ran into limitations with Chrome’s Manifest V3, especially with background service workers and storage constraints. Managing large amounts of cached content efficiently required building a custom storage management layer.

Integrating AI-based clustering with real-time browsing data was another challenge — striking the right balance between performance, accuracy, and responsiveness wasn’t easy.

Accomplishments that we're proud of

We’re proud that Cache-22 can intelligently predict and cache the content users will most likely need before going offline. It delivers a surprisingly fluid experience, even when the internet isn’t available.

We successfully combined multiple technologies — Chrome’s APIs, Claude AI, Bright Data, and Playwright — into a single, cohesive system. Getting all these components to talk to each other reliably in a browser environment was a big milestone.

What we learned

We learned how powerful prediction-based caching can be when combined with AI-driven analysis. It’s not just about saving pages — it’s about understanding what users will want next.

Working with Chrome’s Manifest V3 gave us a deep understanding of how modern browser extensions manage data, events, and permissions. We also gained hands-on experience with scalable page-capture strategies using Playwright and clustering models.

What's next for Cache-22

Enhanced Clustering Models: Integrate more sophisticated clustering algorithms and user preference learning. We might potentially do this by using Chroma DB to take user's most recent search and do semantic search with it over a vector database of all cached websites. If similar websites semantically exist then we don't need to unnecessarily add more websites to the cache
Cross-Device Sync: Allow users to sync their cached content across multiple devices. Also allow users to connect to other offline devices in the proximity via bluetooth and share their portions of the cached internet, creating a decentralized static web of sorts.
Advanced Analytics: Provide insights into browsing patterns and content consumption habits