FeedSift

Inspiration

In an era of information overload, social media feeds are often noisy, overwhelming, and filled with irrelevant content. We envisioned FeedSift as a powerful tool to cut through the clutter, allowing users to reclaim their attention and transform their social media consumption into a truly personalized, focused, and valuable experience. Our goal was to empower users to define what they want to see, not what algorithms think they want.

What it does

FeedSift is an intelligent Chrome extension that acts as your personal content curator for popular social media platforms. Users simply provide a natural language prompt to their FeedSift "agent" (e.g., "Show me the latest breakthroughs in AI and sustainable energy, but filter out celebrity gossip and political commentary"). FeedSift then dynamically filters and re-organizes the content on supported platforms like X (formerly Twitter), Reddit, and YouTube, ensuring you see only what's most relevant to your specified interests, directly within the familiar interface of those apps.

How we built it

FeedSift's architecture comprises two main components: Backend AI Engine: We leveraged the Toolhouse agentic API to power our content analysis. This service takes the user's custom natural language prompt, processes it to understand the core intent and filtering criteria, and then provides signals or scores for content relevance. Frontend Chrome Extension: Built with JavaScript, HTML, and CSS, the Chrome extension is responsible for: Providing a user interface to input and manage their personalization prompts. Communicating with the Toolhouse API backend. Dynamically interacting with the Document Object Model (DOM) of supported social media websites to identify, filter, and re-display content according to the AI's relevance assessment.

Challenges we ran into

The most significant challenge was robust frontend integration and DOM manipulation. Each social media platform (X, Reddit, YouTube, etc.) has a unique, complex, and often dynamically changing DOM structure. Identifying Target Elements: Reliably identifying the correct HTML elements representing individual posts or content items across these varied structures was difficult. Parsing and Extraction: Extracting the necessary information (text, images, author, etc.) from these elements without breaking due to minor site updates required adaptive parsing logic. Seamless Overlay/Filtering: Injecting our customized feed or filtering existing content in a way that felt native and didn't disrupt the user experience or site functionality took considerable iteration and debugging. This reverse-engineering and adaptive scripting consumed a major portion of our development time.

Accomplishments that we're proud of

Functional AI-Powered Filtering: Successfully implementing the core vision of allowing users to define their feed through natural language and seeing it work across multiple platforms. Overcoming DOM Complexity (for key platforms): Developing a flexible-enough approach to parse and manipulate content on complex sites like X and Reddit, demonstrating the viability of our concept. Seamless API Integration: Effectively integrating the Toolhouse agentic API to drive the content understanding and relevance scoring, which formed the "brains" of our operation. Delivering a Proof-of-Concept: Creating a working prototype that tangibly improves the signal-to-noise ratio on social media for the end-user.

What we learned

Advanced DOM Manipulation & Resilience: Gained deep insights into the intricacies of cross-platform DOM manipulation and the necessity of building resilient selectors and parsing logic to handle website updates. Practical AI Application: Learned how to effectively translate user needs into actionable prompts for an AI agent and integrate AI-driven decision-making into a real-world application. Chrome Extension Development: Honed our skills in JavaScript, asynchronous operations, and the specifics of Chrome extension architecture, including content scripts, background services, and inter-component communication. The Importance of Iterative Design: The frontend challenges, in particular, reinforced the value of rapid prototyping, testing, and iterative refinement when dealing with external, uncontrolled environments like live websites. Problem-Solving & Debugging: Enhanced our abilities to debug complex interactions between our extension, third-party APIs, and dynamic web pages.

What's next for FeedSift

We're excited about the potential of FeedSift and have several plans for its future: Expand to Mobile App: Develop native iOS and Android applications to bring personalized feed filtering to users on the go. Broader Platform Support: Integrate with more social media and content platforms (e.g., LinkedIn, Facebook, Instagram, news aggregators). Enhanced Personalization Controls: Introduce features like negative keywords, intensity of filtering, user-defined blocklists/allowlists, and the ability to "train" the agent by liking/disliking sifted content. User Accounts & Syncing: Allow users to create accounts to save their preferences and sync them across devices. Performance Optimization: Continuously refine the DOM interaction logic for speed and efficiency to ensure a smooth user experience.

Built With

• javascript
• toolhouse