Newsference

Newsference is a site dedicated to analyzing and understanding our news and business cycles. Search any topic and get detailed metrics and summaries about how it is being covered.

Comment

Inspiration

We were inspired by ground.news. The website provides articles from different political perspectives but we wanted to add on to that capability. Our site provides articles but it also gives AI-generated summaries for the different political perspectives. We also wanted to reduce echo-chambers by allowing people access to quick info about what the whole political spectrum is saying.

What it does

Search a topic in the news and it will give summaries about the topic from different political views (left, right, and center) and then provide articles from each side.

How we built it

We built it with a HTML and CSS front-end and with a python back-end and htmx linking the two. We also link to the APIs necessary to fetch news and feed the articles into OpenAI.

Challenges we ran into

We ran into challenges first configuring our API responses to ensure that they are relevant to what users typed. We also had difficulties connecting our front-end to the backend using htmx and css styling would get messed up.

Accomplishments that we're proud of

We're proud of the fact that the site acts as a circumnavigation around OpenAI's unreliability regarding current news. It is a much better system to first use NewsAPI and then feed OPenAI that info first, rather than just straight up asking OpenAI.

What we learned

We learned to use htmx to connect a front-end and a back-end. We also learned how to scrape web articles and using the Internet Archive to retrieve inaccessible documents. We also learned how to handle server-side events using text streaming.

What's next for Newference

What's next for Newsference is that there needs to be more prompt engineering to prevent prompt injection when processing news articles. We also need to fine-tune the LLM to recognize political bias more accurately. On top of that, Newsference will also be able provide other data about the articles such as relevancy, date, and quantifying info(if numbers are in the article).

Built With

Share this project:

Updates