Adaptive Air

AdaptiveAIR is an AI wellness dashboard integrating wearable data (Apple Watch, Fitbit,Whoop) for HRV, sleep, recovery insights with an AI coach and cycle-aware women’s health recommendations.

Comment

Inspiration

Three underserved gaps in fitness tech inspired AdaptiveAIR:

  1. Fatigue Blindness — Apps track what you did, not how your body is responding. Without real-time fatigue signals from wearables, users overtrain and plateau.
  2. Disconnected Wearables — Apple Watch, Whoop, Fitbit, Samsung all siloed. No single platform unifies and acts on that data.
  3. The Female Health Gap — Menstrual cycles create profound physiological shifts across four phases that affect energy, strength, and recovery. Zero mainstream fitness platforms dynamically adapt to cycle phase. Women are given the same programming as men, designed from male-centric research.

We built AdaptiveAIR to extend AIR Health's real-time adaptation paradigm from the workout session to the whole-body, whole-cycle wellness journey.

What it does

AdaptiveAIR is an AI-powered adaptive wellness dashboard with three integrated layers:

  • Universal Dashboard — Goal-based workout and dietary recommendations personalized to diet type (vegetarian/non-veg/vegan/eggetarian) and allergies. One-time onboarding, tailored forever.
  • Wearable Intelligence Hub — Connects to 500+ devices (Apple Watch, Fitbit, Garmin, Oura, Whoop, Samsung) via Junction API. Displays real-time HRV, heart rate, sleep score, recovery, SpO2, and respiratory rate with daily/weekly/monthly fatigue graphs.
  • Her Mode — A cycle-phase-adaptive women's health module. Logs periods, predicts next cycle, detects current phase (menstrual/follicular/ovulatory/luteal), and dynamically adapts all exercise, nutrition, and recovery recommendations to hormonal phase.
  • Aira AI Chatbot — Context-aware wellness coach powered by Llama 3.3 70B (Groq). Knows your goal, cycle phase, and HRV. Gives evidence-based, personalized answers.

How we built it

Frontend: React 19, Vite, Tailwind CSS v4, Framer Motion, Recharts, TypeScript (strict mode)

Backend: Hono (lightweight, Bun-native), TypeScript, running on Bun runtime

AI Pipeline: Groq (Llama 3.3 70B) as primary, with Claude and Gemini as fallbacks. Backend proxies all AI calls to keep API keys secure.

Wearable Integration: Junction/Vital unified API with OAuth widget supporting 500+ devices. Real biometric data flows through REST endpoints for sleep, activity, heart rate, and body metrics.

Auth: Firebase Authentication with Google Sign-In. Per-user data isolation using Firebase UID-scoped localStorage.

Architecture: Monorepo with Bun workspaces (frontend/ + backend/). Vite dev server proxies /api/* to Hono backend. Concurrent dev script runs both with one command.

Challenges we faced

  • No wearable has a web API for direct browser access. Apple HealthKit is iOS-only, Google Health Connect is Android-only. We solved this by integrating Junction's unified API which handles OAuth for cloud-connected devices (Fitbit, Garmin, Oura, Whoop) and provides mobile SDKs for Apple Health/Samsung.
  • Every AI API ran out of credits or quota. Anthropic had $0 balance, Gemini free tier quota was 0. We built a 3-tier fallback chain (Groq → Claude → Gemini → local knowledge base) so the AI chatbot always responds.
  • Fitbit OAuth redirect wiped React state. The page reload on OAuth callback lost the user's goal selection. Fixed by persisting critical state to localStorage scoped to Firebase UID.
  • Women's health data is structurally different from fitness data. Cycle phases affect every recommendation differently. We built a research-backed phase recommendation engine sourced from PMC10251302, PMC6257992, and PubMed 35471634.

What we learned

  • How real wearable data pipelines work (OAuth → cloud API → normalized schema → dashboard)
  • The science behind cycle-phase training periodization and hormonal nutrition
  • Building production-grade AI fallback chains for reliability
  • Firebase Auth integration with user-scoped state management
  • Monorepo architecture with Bun workspaces

What's next for AdaptiveAIR

  • Expo React Native companion app to read Apple HealthKit directly from iPhone (enabling real-time Nothing Watch 2 data)
  • Groq-powered AI that generates personalized weekly meal plans based on cycle phase + dietary preferences
  • Longitudinal wellness fingerprint: HRV trends vs cycle phase overlay, sleep quality vs workout performance correlation

Built With

Share this project:

Updates