WellAware AI
WellAware AI
AI that connects your wearable's missing dots
About the Project
WellAware AI is a personal wellness intelligence platform that bridges the gap between wearable data and actionable understanding. Over 100 million people wear health trackers, yet the vast majority never get meaningful insights beyond daily step counts and sleep scores. WellAware connects your Oura Ring biometrics — sleep stages, HRV, readiness, temperature — with your Google Calendar schedule to surface cross-domain patterns that emerge over weeks and months.
The platform's AI agent doesn't just answer questions about last night's sleep. It remembers your patterns across sessions, tracks your wellness goals, and proactively warns you when it detects familiar precursors — like the specific combination of travel, meeting density, and declining HRV that preceded your last illness.
As sole founder and technical architect, I built WellAware from concept to closed beta, drawing on my neuroscience PhD and decade of experience in biomedical data science and ML (NeuroSky, Aspen Neuroscience, Cruise, AWS)
The core innovation is a multi-layer memory architecture: the AI maintains short-term conversational context, medium-term session patterns, and long-term persistent memory with semantic search — so it genuinely learns you over time, not just your data. The platform also features advanced temporal reasoning, allowing natural-language queries across arbitrary time ranges — something most AI assistants fundamentally lack.
Project Execution
The technical stack reflects the complexity of building a stateful, temporally-aware AI agent at production quality. The backend runs on GCP Cloud Run with a LangGraph-based agent orchestration layer, using BigQuery and pgvector for analytics and semantic memory storage. Firebase handles authentication, and the frontend is built in Streamlit for rapid iteration during beta.
Infrastructure includes full OpenTelemetry instrumentation for observability, DNS managed through Cloudflare for deliverability, and a marketing automation pipeline connecting Loops (email), Zapier (workflows), and a custom waitlist system. The Oura Ring integration uses their OAuth2 API, with plans to expand via aggregator APIs (Vital, Terra) to support Whoop, Garmin, and Apple Health.
The hardest engineering challenges weren't the integrations themselves but the temporal reasoning layer and the memory architecture. Most LLM-based systems have no concept of "last Tuesday" vs. "two weeks ago" — I built an explicit temporal context processing system that grounds every query in real calendar time. The multi-layer memory system required careful design to balance relevance, privacy, and retrieval speed across thousands of stored observations.
The beta launch involved migrating the waitlist to Loops, setting up personalized onboarding email sequences, and instrumenting the entire user journey for telemetry. Currently accepting 100 early testers with plans to expand wearable integrations and introduce scheduled automated reports.
