AI Newshound assists journalists in discovering non-obvious AI research topics with increased traction and media attention, and visualizes these trending subjects in a user-friendly manner.
We interviewed various types of journalists to identify the primary challenges in story writing or television news communication. We discovered that journalists strive to stay ahead of the obvious, seeking compelling topics to write and research. Additionally, as emphasized by journalists, their objective is to “communicate the truth”, sourcing accurate evidence for their stories as swiftly as possible. For this reason we found great opportunities to help journalists to identify new stories and validate and visualize different sources to validate their hypothesis.
A research co-pilot that helps journalists to find “the needle in the haystack” to uncover news-worthy insights.
Mockup the site – future vision
When a user goes to AI-Newshound, they can search for a specific topic and visualize trends. We ingest articles, convert them into embeddings, and generate a 2D visualization, which is then plotted and rendered with Streamlit.
Co-creating with the user:
Will Knight, journalist at Wired, joined since the beginning of the hack taking the role of the User-Lead.
We brainstormed a list of potential ideas based on Will’s experience and user research with others from the week before. Trends stood out as an important topic, and after brainstorming 25+ ideas, we moved forward building Weekly reports on top AI papers.
We started working, but it wasn’t long before our first pivot. One of our members came back from lunch talking excitedly about Pathfinder, a tool to visualize Arxiv papers. Inspired, we showed it to Will, who said that the previous tool made something he’d do himself faster, but this new direction opened up a way to do research he’s never had access to before.
Technical development:
We started with the goal of putting an MVP 0 together, we broke into groups and each of us pulled data for “AI Agents” from a different API source.