NewsBro

Inspiration

Everyone hates fake news. AI can help us solve that problem.

What it does

Newsbro consumes news articles for a specific headline and presents five major unbiased takeaways along with the sources for these takeaways.

How we built it

Using LexRank, Meta's BART, and Google's universal sentence encoding for news processing. LexRank was used to extract the most important sentences from long articles, reducing the number of tokens per article. BART was then used to summarize the outputs from LexRank, and generate bullet points. These bullet points were then fed into the universal sentence encoder to output the five most relevant takeaways for a given news headline. The app backend runs on a Flask server, and the frontend using Next.JS.

Challenges we ran into

Reducing the number of tokens for long text inputs without losing meaningful data. Matching sentences based on how similar they are to generate the takeaways Integrating this functionality with a smooth user experience, making it easy for everyone to use.

Accomplishments that we're proud of

Designing the end-to-end pipelines and implementing it all in less than one day. Coming up with plans to expand on this and improve in the future. Using white-paper research to inform design decisions.

What we learned

There is a huge open source AI community that is ready to help. You can build amazing tools with just a day's work. We should participate in more hackathons.