-
-
Landing Page
-
On submission
-
Inspiration
Every Team USA fan has wondered — "what sport would I have competed in?" With 120 years of Olympic and Paralympic history, we have the data to actually answer that question through body type analysis and AI storytelling.
What It Does
Find Your Team USA Twin is a "Digital Mirror" — a fan engagement tool that matches your biometric profile (height, weight, age) to one of 6 historically-grounded Team USA athlete archetypes, with equal analytical depth for both Olympic and Paralympic sports.
Fans input their physical traits and sport interests. The app:
- Computes normalized Euclidean distance against real Team USA athlete centroids derived from 14,214 historical records
- Ranks all 6 archetypes with a scored comparison
- Generates a personalized narrative using Gemini 2.5 Flash — historically grounded, no athlete names or likenesses
- Shows matching Olympic AND Paralympic sports with equal depth
The 6 archetypes: ⚡ Sprint Power Unit, 🏊 Aquatic Specialist, 🤸 Kinetic Artist, 🏋️ Power Explosive, 🚴 Lean Aerobic Engine, 🏀 Court & Field Athlete
How We Built It
Matching Engine: Pandas loads and filters the dataset to USA-only records (NOC == 'USA'), drops all PII columns (Name, ID, Event), and computes archetype centroids from real biometric means. The matcher normalizes height, weight, age, and BMI using Team USA population statistics, then applies Euclidean distance scoring with a sport-interest bonus weighting.
Gemini 2.5 Flash: Called server-side via the Gemini API to generate personalized narratives. The prompt enforces: no athlete NIL, conditional phrasing ("could align with", "may excel in"), and dedicated Paralympic analytical depth in every response.
Google Cloud Run: FastAPI backend deployed as a containerized service. Stateless, auto-scaling, HTTPS out of the box.
Frontend: Animated vanilla HTML/CSS/JS with particle field, orbital loading animation, radial match gauge, and staggered reveal animations.
Data Compliance
- Dataset filtered to
NOC == 'USA'only — 14,214 records across 54 sports - Name, ID, and Event columns dropped at load time (no NIL)
- Only Medal placement used — no finish times or scores
- No IOC branding anywhere in the application
- Conditional phrasing enforced throughout: "could align with" not "will"
- Paralympic sports shown with equal analytical depth for all 6 archetypes
Challenges
Getting the sport interest weighting right was the key challenge — pure biometric distance would always favor the nearest body-type cluster regardless of what sport a fan actually loves. The solution was scaling the biometric distance by 0.4 and applying a 2.0 std-dev bonus per matching sport, so a fan who picks Archery gets Kinetic Artist even if their BMI skews slightly toward Power Explosive.
What We Learned
120 years of Team USA data tells a remarkable story about how body types cluster across disciplines — and how those same physical profiles translate directly to Paralympic competition. The data proves that athletic excellence is truly universal.
Log in or sign up for Devpost to join the conversation.