StepSquad - Move Together, Win Together

Team up, sync your smartwatch, and turn every step into a team challenge! StepSquad connects Garmin, Fitbit & phones to power global step competitions and fair AI-driven leaderboards.

Comment

StepSquad – Move Together, Win Together

Inspiration

Our company recently launched a corporate wellness challenge using another step-tracking app — but we quickly discovered several issues: inaccurate syncing, missing charts, and even ways to cheat the rankings.

So we decided to build StepSquad, a cleaner, fairer, data-driven platform designed for real-world team challenges.

We're a small team of two engineers who love both fitness and software craftsmanship — and wanted to prove that an elegant, transparent solution could be built fast using Google Cloud Run and AI.

What it does

StepSquad connects with Garmin and Fitbit devices via OAuth to collect verified step data during competitions.

Each participant joins a team, and both individual and team leaderboards update in real-time.

AI-powered agents built with Google ADK analyze the data using Gemini 2.5 Flash to detect anomalies (e.g., unrealistic spikes or suspicious patterns).

The result: a fun, trustworthy competition experience for companies and communities.

How we built it

The platform runs on a modern, cloud-native microservices architecture with 4 Cloud Run services:

  • Frontend: React 18 + Vite + TypeScript (web dashboard) deployed on Cloud Run with custom domain www.stepsquad.club

  • Backend API: FastAPI (Python 3.11) served on Cloud Run with custom domain api.stepsquad.club

  • Workers Service: Background worker on Cloud Run for syncing step data from linked devices via Pub/Sub

  • AI Agents Service: Google ADK multi-agent system on Cloud Run with Sync Agent and Fairness Agent powered by Gemini 2.5 Flash

  • Data: Firestore (user & competition data) + BigQuery (analytics)

  • Messaging: Pub/Sub event pipeline between ingestion API and background workers

  • Infrastructure: Automated CI/CD via GitHub Actions, custom domains configured

Architecture highlights:

flowchart LR
    U[User] --> Web[React Web App<br/>Cloud Run]
    Web --> API[FastAPI Backend<br/>Cloud Run]
    API --> Pub[Pub/Sub Topic<br/>steps.ingest]
    Pub --> W[Worker Service<br/>Cloud Run]
    W --> F[Firestore]
    W --> BQ[BigQuery]
    F --> API
    API --> Agents[AI Agents Service<br/>Cloud Run<br/>Google ADK]
    Agents --> G[Gemini 2.5 Flash]
    F --> Web
    Device[Garmin/Fitbit] --> API

Challenges we ran into

  • Integrating multiple health APIs (Garmin, Fitbit) with different OAuth flows and rate limits. Garmin uses OAuth 1.0a while Fitbit uses OAuth 2.0, requiring different handling.

  • Multi-agent orchestration with Google ADK — ensuring agents communicate effectively while maintaining separation of concerns.

  • Handling timezone alignment for competitions and daily resets across different regions.

  • Designing fairness detection logic that's accurate but not too strict — using AI to balance human judgment with automated flagging.

  • Keeping the infrastructure lean for hackathon speed while staying production-ready with proper authentication, error handling, and monitoring.

  • Deploying 4 separate Cloud Run services with proper dependencies, environment variables, and custom domains.

Accomplishments that we're proud of

  • End-to-end working prototype with real cloud data ingestion, live leaderboards, and custom domains (www.stepsquad.club and api.stepsquad.club).

  • Multi-agent AI system built with Google ADK meeting all hackathon requirements for the AI Agents category.

  • 4 Cloud Run services working together seamlessly: frontend, backend, workers, and AI agents.

  • Production-ready deployment with automated CI/CD via GitHub Actions and Firebase Authentication.

  • Modern, responsive UI built with React, TypeScript, and Tailwind CSS that looks great in any mode.

  • Building it all as a team of two while working full-time elsewhere. 💪

What we learned

We learned how to combine serverless microservices on Cloud Run with modern web frameworks in a matter of days.

We mastered Google ADK for building multi-agent systems and saw how Gemini 2.5 Flash can provide intelligent analysis at scale.

We discovered that wellness + gamification can be done ethically and transparently, with technology supporting — not faking — real activity.

We learned the importance of proper architecture — separating concerns across services while maintaining clean communication patterns.

What's next for StepSquad - Move Together, Win Together

  • Mobile app: Build a Flutter mobile app with direct HealthKit and Health Connect sync.

  • Fairness engine expansion: Enhance anomaly detection with more sophisticated AI-based scoring and pattern recognition.

  • Company dashboards: Add analytics with charts, heatmaps, and trends over time using BigQuery.

  • Open challenges: Enable public leaderboards for cities, events, and charity walks.

  • Smartwatch companion app: Quick status and daily progress visualization.

  • Additional device integrations: Support for more fitness trackers and health platforms.

Together, we believe every step — whether 5,000 or 50,000 — deserves to count, fairly.

Built With

  • apple-health)-ai-&-automation:-google-agent-development-kit-(adk)-?-fairness-&-sync-agents-devops-&-ci/cd:-docker
  • bigquery-messaging-&-events:-pub/sub-infrastructure-as-code:-terraform-authentication:-oauth2-(garmin
  • cloud-build
  • cloud-storage-data-&-analytics:-firestore
  • dart-frameworks:-fastapi-(backend)
  • fitbit
  • flutter-(mobile)-cloud-platform:-google-cloud-run
  • github-actions-design-&-assets:-figma
  • google-fit
  • languages:-python
  • lucide-icons
  • make
  • mermaid
  • pnpm
  • react-+-vite-(web)
  • tailwind-css-other-tools:-uv-(python-env-manager)
  • typescript
Share this project:

Updates