FocusFeed

Our simplistic landing page
After logging in, users take a quick and fully customizable survey about their newsletter preferences so they receive tailored news.
An example of the email you receive with your newsletter for the day.
The newsletters cover the topics you choose, with summaries given in the tone you want. The source of the article is also given.
Our full tech walkthrough, including how different technologies interact to create the perfect end-to-end solution.

Inspiration

In an age of information overload, it’s challenging to keep up with all the content that matters. We wanted to create a solution that simplifies the process, providing users with bite-sized, AI-summarized updates from the topics and channels they care about most. The idea for FocusFeed was sparked when one of our team members heard this problem firsthand from a friend’s parent, who expressed frustration about the overwhelming amount of content they had to sift through daily. They suggested, "Why don’t we harness the power of AI to make these updates personalized?" This inspired us to blend personalization and AI-powered summarization into a seamless, efficient experience—helping users stay informed without spending hours scrolling.

What it does

FocusFeed is a personalized newsletter service that curates and summarizes content based on user preferences. Users can sign up, select their favorite topics or channels, the tone they want their newsletters in, and set their desired newsletter frequency. We then scrape the web and our fine-tuned GenAI model aggregates new content from different sources and summarizes it into concise, digestible updates, delivered directly to users' inboxes. It’s like having a personal news assistant that brings the most relevant updates without the noise.

How we built it

We built FocusFeed using a modern tech stack:

Backend: Developed with Python and NodeJS to handle API integrations, user management, and content aggregation. Frontend: Built using Next.js/React.js for a dynamic, responsive user experience, styled with Tailwind CSS for fast and sleek UI development. Database: GCP Firebase handles user data, content storage, and preferences, ensuring scalability and real-time capabilities. AI Summarization: We leverage Perplexity/Llama 3.1 models for accurate and efficient summarization, ensuring that users receive only the most concise, high-quality summaries of their selected content. Email Service: We used a simple SMTP library for sending newsletters. Google Cloud Functions manage the scheduling and automation of daily/weekly newsletters, ensuring timely and reliable content delivery.

Challenges we ran into

One of the primary challenges was optimizing the AI summarization to maintain context while condensing information into short, engaging summaries. We also faced difficulties with integrating multiple APIs for content aggregation, as they varied in format and reliability. Ensuring the newsletter delivery was timely and avoided spam filters required thorough testing and adjustments to our email service implementation.

Accomplishments that we're proud of

We’re proud of successfully integrating AI summarization into a user-friendly platform, delivering concise and valuable updates tailored to individual preferences. Building a system that automates the entire workflow—from content aggregation to email delivery—was a significant achievement, especially within a limited timeframe. The intuitive and polished UI we designed using Next.js and Tailwind also stands out as a highlight. We were able to solve a real problem that people want a solution for, using LLMs to personalize your daily news.

What we learned

Throughout this project, we deepened our understanding of AI models, particularly how to fine-tune them for summarization tasks while balancing context and brevity, as well as using system prompts to tailor the responses of GenAI. Because we also needed to send different information to different users at different times, we also gained experience with Firebase and Cloud Function orchestration, ensuring a seamless and scalable backend solution. We were able to asynchronously send information with the Cloud Functions to solve this. Additionally, working with real-time content aggregation APIs taught us how to handle inconsistencies and data standardization.

What's next for FocusFeed

Moving forward, we plan to enhance FocusFeed by incorporating sentiment analysis, allowing users to filter content based on positive, neutral, or negative tones. We also aim to expand content sources and provide multi-language support, making the platform accessible to a broader audience. We are also planning to create a Reinforcement Learning based Recommendation Algorithm so that users can see "Suggested Content". Long-term, we envision adding personalized audio versions of newsletters, so users can listen on the go, transforming FocusFeed into a comprehensive, cross-channel media digest tailored to every user's lifestyle. We want this to be a "one stop shop" for everyone's daily news, personalized for your interests and needs.