WellNest

Inspiration

We live far away from our parents and grandparents, and constantly worry about their well-being and daily routines. We wanted to create something that gives us and other families real peace of mind, without invading privacy. WellNest lets us know they're safe, active, and following their normal patterns, even if we can't be there in person.

What it does

WellNest uses non-intrusive sensors and advanced AI to monitor loved ones' daily routines in their own homes. Tracks key routines: wake-up time, kitchen activity, bathroom visits, bedtime, and more Detects anomalies in real time—missed kitchen visits, prolonged inactivity, unusual bathroom use Uses semantic search so families can ask questions like "Show all days with late wake-ups" or "Has Mom’s routine changed this week?" Provides actionable alerts and natural-language LLM summaries for caregivers, making sense of thousands of sensor events at a glance. Keeps data private - no cameras, just peace of mind

How we built it

Backend: Python (FastAPI) on AWS EKS (Kubernetes), MongoDB for storage Simulator: Custom event generator creates realistic sensor data for dozens of households AI/ML: NVIDIA NIM for retrieval embedding (nv-embedqa-e5-v5), Qdrant for vector search, llama-3 1-nemotron-nano-8B-v1 for LLM summaries and anomaly explanations Frontend: React, Typescript, Tailwind dashboard with live activity feeds, routine cards, trend charts, and semantic search Integration: Real-time alerts (WebSocket), dashboard search all unified in Python

Challenges we ran into

Making routine analysis general enough for any home, but specific enough to catch subtle behavioral changes Simulating enough diverse sensor data to stress-test anomaly detection and semantic search Integrating AI services (embedding search, LLM summaries) with EKS in a way that feels effortless for users Designing a dashboard that's simple for families, but exposes all the depth of the underlying AI

Accomplishments that we're proud of

Built a truly end-to-end AI system—from data ingestion to actionable insight on cloud infrastructure (AWS EKS) Working semantic search: caregivers can ask questions in plain English and get instant, meaningful results AI-generated natural-language daily summaries and anomaly explanations humanizing complex sensor data Live demo with simulated and test data Privacy-first approach: no cameras or mics just sensors and smart analytics

What we learned

How powerful and accessible modern AI models (NIM, Llama-3) are for practical, peace-of-mind problems The value of personalized, explainable alerts not just "motion detected!" spam How to leverage vector search and embeddings to connect the dots in daily life, not just plain text That end users value clarity and control a simple, beautiful dashboard beats a technical console

What's next for WellNest

Integrate with more smart home systems (motion/door sensors, voice assistants) Add multi-household and multi-resident analytics; enable "family" notifications for multiple caregivers Launch pilot tests with real families and care agencies for feedback Expand anomaly models with more sophisticated trends and personalized recommendations Offer WellNest as a subscription service to families, elder care agencies, senior living communities Improve accessibility (mobile app, SMS alerts) and add more natural interactions with dashboard AI Continue to champion privacy-preserving, ethical monitoring for older adults