BALANCE BRIEF

Article screen
Homepage
Trending screen

NOTE: Our backend needed to restart during recoding. That is the reasoning for the weird cut.

Inspiration

We all read the news daily but get pushed to think one way without even realizing. In today's politically polarized world, it's hard to get the full story. We wanted to create an app that would understand bias and read/processes articles for you.

What it does

It uses a study by AllSides that maps popular news outlets with their political alignments. We then read in articles from these news sites, add their biases, and then use this to match our search query to an unbiased summary of a combination of the matches. The app will then display this summary to the user, briefing them on the concept they searched, while also listing 3 sources, each with a different major implicit bias, for the user to further explore. This is all driven with AI, as we explain shortly.

How we built it

We build this using some of the newest and most advanced technologies. We used Angular for the front end with a custom-designed UI library. To handle the querying and actual searching, we transformed thousands of articles into weighted vectors using Cohere and stored them in a Qdrant table. We then do the same procedure for the user's query, except that this is not stored in Qdrant. Instead, this vectorized query is compared with the vectorized articles in the Qdrant database using cosine similarity. The top matches from each major bias (left, right, and central) are then gathered, put together, and summerized using Cohere generative AI functions. The metadata stored along with the vectorized articles in the Qdrant array is then accessed and displayed to the user as appropriate. Lastly, we created a custom Python web Crawler to create an article database from many articles across the web.

Challenges we ran into

We bit off a lot more than we could chew, it was hard to get these powerful technologies to work together. Our app had many processes that were all built on top of each other.

Accomplishments that we're proud of

We worked as a team! Everyone did their part, we even had someone that does not know how to code and he was able to very helpful. Everyone was happy, and that's all that really matters. We are proud that we were able to work with complex tools that were brand new to all of us and come up with a well-functioning prototype. The tools that we used are the absolute cutting edge of generative AI, as Cohere v3 was literally released 2 days ago.

What we learned

We learned the core concepts of how to interact and program with a large language model. We also learned how to work with these new generative AI features and combine them in a way that has not previously been done to achieve a new result. And of course, we learned a lot about some programming concepts but more importantly, we learned to work together in our own ways but combine our content.

What's next for BALANCE BRIEF

We want to build this to be a more fleshed-out app. We want to improve our web crawler so that we can schedule it to run when we desire so that we can automatically gather and categorize articles using bias to only further improve our product. What we have is super powerful. News should inform you and not tell you how to feel, that's for you to decide. We want to have this technology more accessible so people can truly be informed.