Team USA Hubs

Support your hometown! What makes them unique? Find, analyze, share on your favorite USA sports hub with high quality data and Gemini's assistance.

Comment

Inspiration

Every four years, we cheer for Team USA without really knowing where these athletes come from. A gymnast from small-town Iowa, a swimmer from coastal California, a wheelchair racer from Illinois — each has a hometown story that gets lost in the medal count.

We wanted leverage the power of Google platform and AI to flip the lens and ask: what makes a hometown unique and successful for Athletes? We want to provide a clean, visual way that supports both simple and in depth analysis for different audiences.

What it does

Team USA Hometown Hubs maps every Olympic and Paralympic athlete back to their hometown. Users can explore the data in various way.

  • By Hub — k-means clusters that reveal regions state lines don't show
  • Explore — visualize the trend and demographics by time series and sports Freeform chat — Ask any questions with help of AI to answer

The app also have additional functionalities like lets users compare any two hubs side-by-side, analyze how the athletic and geographical culture helped and generate shareable infographic card.

How we built it

  • App — main web app built on AI Studio
  • Data Scraping — built a scraper in AI Studio
  • Data Cleaning — Gemini API for smart cleaning of messy roster data
  • Storage — Firebase for the cleaned dataset and as a cache for hub comparisons
  • Geocoding — Google Maps API to turn hometowns into coordinates
  • Clustering — k-means on coordinates to form hubs, balanced across Olympic and Paralympic athletes
  • Querying — GCP-hosted reasoning translates natural-language questions into structured queries
  • Image gen — Nano Banana generates infographic cards

Challenges we ran into

  • Hometown data is messy: "born in," "trained in," and "represents" cities don't always match
  • Geocoding small towns is ambiguous (there are a lot of Springfields)
  • Choosing the right number of k-means clusters is a storytelling choice
  • Raw counts skewed toward Olympic athletes, so we had to weight Paralympic representation explicitly
  • LLM costs forced us to add Firebase caching

Accomplishments that we're proud of

  • Clean data with verified results
  • Clear visualization across different dimensions
  • AI tool offering flexibility
  • Equal representation of Olympics and Paralympics athletes

What we learned

  • Power of the AI made the non-trivial/impossible task achievable.
  • Data storytelling lives or dies on the cleaning step
  • Paralympic data deserves equal billing by design. The respect
  • Cost control with caching for LLM use

What's next for Team USA Hometown Hubs

  • Finish generating for most common cases and rollout to public
  • Add more data regarding the areas
  • Add 2026 candidates in the list with special section

Built With

Share this project:

Updates