Team USA x Google Cloud
Public Team USA athlete records since 1904, projected into one living map.
Olympic records render as filled circles. Paralympic records render as rings.
Method
Public Olympic rows, public Team USA profile context, and IPC Paralympic/classification context are normalized into aggregate records. Identifying fields never ship to the browser.
The deployed artifacts contain 18,356 public Team USA records, 8 aggregate zones, and 14,205 body-comparison records. The current strongest Paralympic-ratio zone is Power-to-Weight: 5,474 Olympic records and 46 Paralympic records.
/api/match scores the session profile against generated zones. Gemini on Vertex AI receives only
aggregate context and writes optional explanatory copy; it does not choose the match.
Body Atlas never returns an athlete name, performance prediction, sport prescription, diagnosis, or Paralympic classification assignment. Safety checks rewrite deterministic language into conditional language.
Paralympic biometric data is sparse in public sources. Body Atlas gives every available Paralympic record the same analytical treatment as Olympic records, but it does not pretend missing fields exist.
The demo runs on Cloud Run. Vertex AI Gemini is the controlled narrative layer. Cloud Build packages each source deploy. The app uses service-account auth, not an AI Studio API key.
Story
Every exported point is an anonymous public Team USA record, projected by aggregate body archetype.
Each region contains athletes who share public body-pattern context, not identities.
The highest-share zone still contains far more Olympic records than Paralympic records. The atlas surfaces that gap instead of hiding it.
The time-warp mode lets the atlas rearrange decade by decade around the user's anchored point.
The full method page documents the public-data pipeline, deterministic matcher, Vertex AI Gemini narrative layer, and safety auditor.