Inspiration
While many see the US as a single entity, its true strength is a mosaic of regional excellence. Hometown Heroes explores the diverse environments that inspire Team USA to compete in the Olympic and Paralympic Games, proving that our national success is rooted in the unique landscapes of our hometowns.
What it does
Hometown Heroes is an interactive discovery engine that maps the geographic foundations of Team USA excellence. It transforms raw athlete data into a high-performance visual dashboard where fans can explore the "Success Hubs" that forge world-class talent.
How I built it
My goal was to bridge raw data and fan engagement using a Google Cloud pipeline. First, I used Gemini 2.5 Flash Lite to extract structured data from Team USA Athlete PDFs, which was then enriched with geographic context via Google Maps APIs. My architecture utilizes BigQuery for heavy analysis and Cloud Firestore as a low-latency serving layer for an instant UI experience. I leveraged Gemini 2.5 Pro and Imagen 3.0 to generate compliant narratives and optimized community visuals. Finally, the system was deployed on Google Cloud Run to ensure a secure and scalable environment.
Challenges I ran into
The first suggestion Gemini makes isn't always right; I learned that when I had to migrate many generated images from BigQuery to GCS. In hindsight, I should have formed an architectural plan, revisiting to make adjustments before generating large amounts of data. Google Cloud services like GCS and Firestore were new to me, so I didn't quite understand the limitations of BigQuery. The same could be said for UI/UX design; having initial designs would have let me avoid making micro-adjustments to the frontend.
Accomplishments that I'm proud of
Honestly I'm just happy everything fit together and ran into cloud with generally little debugging, especially since I vibe coded this in a few days' time. I'm most proud of lazy loading more non-critical modules and write-through caching AI generated content for improved user experience.
What I learned
Gemini is definitely powerful at supporting 0->1 development, from fast prototyping to thorough validation to finetuned deployments. I also enjoyed the exposure to the Google Cloud environment, where my skills in more common systems architecture translated well to Google's inhouse tools.
What's next for Hometown Heroes
Integrating participation years, regional climate, and elevation could lead to some interesting observations about how the influences on athletes' hometowns. Creating interfaces to data on individual sports, specific Olympic and Paralympic Games, and states/regions would increase the value of the dataset. Providing an avenue for fan interaction would make the engine more engaging; they could show their support by "cheering" on their hometown, or generating a "postcard" through Imagen to share.
Built With
- cloudrun
- db-dtypes
- fastapi
- gemini
- google-auth
- google-cloud
- google-cloud-aiplatform
- google-cloud-bigquery
- google-cloud-firestore
- google-maps
- gunicorn
- imagen
- jinja
- pandas
- pillow
- pydantic
- pypdf
- python
- python-dotenv
- python-multipart
- requests
- tailwindcss
- uvicorn
- vertex
Log in or sign up for Devpost to join the conversation.