Inspiration
People are busy. Not everyone has time to stay updated with the latest news in their interests, whether it be sports, tech, politics, or more. That’s why entertainment has shifted majorly towards short snippets of media, such as tik tok, youtube shorts, instagram reels, etc. However, the information from those sources can’t be particularly trusted, as at the end of the day, they’re forms of social media. We don’t have time to catch up on real news in the same way that we can watch these shorts of entertainment online.
What it does
When users first sign up for Bites, they are prompted to select from a wide array of interests, ranging from technology, health, and science to entertainment, sports, and more niche topics. Once their interests are set, Bites curates a personalized feed in the form of short highlights, ensuring that users stay informed about the latest developments in the areas they care about most, while also delivering accurate and well-sourced information. In addition to the curated news feed, Bites takes personalization a step further by using the selected interests to generate a custom daily podcast for each user. This podcast is a short, digestible format designed to fit into users' busy schedules, providing them with a quick yet comprehensive overview of the latest headlines and trends in their selected topics.
How we built it
We developed the front-end and back-end of Bites simultaneously to satisfy the time constraint. For the front-end, we started by planning the UI in Figma. Although we continued to tweak the design throughout the project, we began translating the ideas into code after we had a rough draft. The framework we chose to use is Flutter, for its performance and variety of features. For the backend, we coded in python, and utilized serpapi to scrape resources from the internet, and OpenAI’s gpt 4 in order to detect truthful information and generate the podcasts. To train each model to detect truthful information, we did 3 shot testing, providing it with three examples with explanations as to why a given article excerpt could be deemed accurate or inaccurate. After these models were trained, the final decision on whether a source was accurate or not was made based on the decision of two LLM’s debating with each other, until an agreement was made. Finally, we used the distinguished truthful key points to generate the podcast.
Challenges we ran into
One of the main challenges we ran into was ensuring that the information provided through Bites was both informative and trustworthy while also keeping it as engaging as popular social media platforms like Tiktok, Instagram, and Youtube. In order to do this, we had to strike the right balance between delivering short, digestible content while also maintaining accuracy and credibility. We developed a self-consistency reasoning algorithm which utilizes multiple LLMs to cross-verify results and ensure consistency of the content. Another issue we ran into was curating personalized news feeds (podcasts).
Accomplishments that we're proud of
We are proud of creating a highly personalized experience that tailors news feeds and custom daily podcasts based on each user’s preferences. By leveraging advanced machine learning algorithms and real-time data processing, we’ve built a dynamic system that adapts as the user's interests change. Additionally, we successfully combined the need for high-quality, credible information with the digestible, short-form style popularized by social media.
What we learned
Throughout the development of Bites, we gained valuable experience in several key areas. First, we deepened our understanding of Flutter, using it to build a responsive mobile application. This framework allowed us to efficiently design a user-friendly interface that meets the needs of users. We gained hands-on experience in LLM models. This was crucial for not only implementing the accuracy checking, but also tailoring the podcasts towards the users interests.
What's next for Bites
Next, Bites aims to utilize a MongoDB database to scale the application for multiple users, enabling a more dynamic and responsive experience. By open-sourcing this database, the platform can provide researchers with instructions for usage, fostering collaboration and innovation. Additionally, Bites will develop a modernized version of Reddit by employing an algorithm that tackles Reddit's current issues—such as inaccurate moderation and poor content—ensuring higher-quality and more reliable information is presented to users.
Log in or sign up for Devpost to join the conversation.