Team USA Analyst Agent

Evidence-bound Gemini-ready analyst agent that turns public Team USA event data into fan dashboards, editorial briefs, and inclusive Olympic/Paralympic storylines.

Comment

Inspiration

Fans want context before, during, and after marquee events: who to watch, what the technical storyline is, and how to understand Olympic and Paralympic sports quickly. Team USA storytelling also needs to stay evidence-bound, inclusive, and easy for editors or fan apps to reuse.

What it does

Team USA Analyst Agent loads public-style event and athlete profile data, scores story priority, and generates:

  • a Markdown analyst brief for editors,
  • dashboard-ready JSON for a fan app,
  • an evidence-bounded Gemini prompt template that only answers from supplied facts.

The current demo uses synthetic sample data so reviewers can run it safely without credentials or private athlete data.

How we built it

The MVP is a Python standard-library agent core with deterministic scoring, sample event data, unit tests, generated dashboard artifacts, and a GitHub Pages landing page. The architecture is Gemini-ready: the repo includes a prompt boundary and adapter point for Gemini / Vertex AI grounding while avoiding committed secrets.

Accomplishments that we're proud of

  • Runnable MVP with no external secrets or paid services required.
  • Evidence-bound insight generation instead of unsupported sports claims.
  • Inclusive Olympic/Paralympic storyline support.
  • Public proof package: landing page, source repo, release ZIP, CI run, and unlisted demo video.

What we learned

The most important product boundary is not just generating exciting fan copy; it is keeping every claim tied to supplied evidence so the output can be trusted by editors, athletes, and fans.

What's next

  • Add a Gemini API adapter and Vertex AI deployment path.
  • Ingest live public event schedules and public Team USA profile sources.
  • Add interactive dashboard filters, accessibility summaries, and multilingual fan modes.
  • Add stronger evaluation for citation coverage and unsupported-claim rejection.

Reviewer links

Built With

Share this project:

Updates