Nomi
An AI-powered home companion for elder care and daily safety
Interface mockup or early screenshot
photo / screenshot coming soon
Why I built this
We wanted to build something that could make independent living safer and less stressful for older adults and the people caring for them. My own grandmother fell down the stairs in India and we had no idea until she was able to get to her phone. A lot of elder-care tools out there are reactive, fragmented, or difficult to use, and they often miss the everyday patterns that signal when something is off. I was interested in creating something that could preserve privacy WHILE continuously observing home activity, connecting sensor data, and gather meaningful conclusions before small issues become bigger problems.
Overview
Nomi is an AI-powered elder-care companion that uses a network of sensors and intelligent agents to monitor daily routines, detect unusual patterns, and provide caregivers with useful updates. It’s designed to be proactive, not just reactive - catching small issues before they become emergencies, while respecting the user’s privacy and autonomy.
The Problem
What's broken
Many older adults want to live independently, but families often have limited visibility into how they are actually doing day to day. Existing solutions are often expensive, intrusive, or too narrow in scope. A fall button or camera alone does not tell the full story.
The Solution
What I built
We built an AI-assisted monitoring system that connects sensor data from the home into an intelligent care platform. Nomi tracks patterns across multiple devices, identifies irregular behavior, and helps caregivers understand when something may need attention. Rather than flooding users with raw data, the system is designed to interpret signals and turn them into useful alerts, summaries, and insights. It’s much more than a fall detector - it’s a daily companion that learns the rhythms of the home and looks out for the people in it.
Technical Approach
Nomi uses a multi-sensor architecture paired with AI agents to process home activity data and generate context-aware outputs. We turned low-cost sensors on an ESP32 into caregiver-ready insights by streaming data to DynamoDB, fusing it in a FastAPI service, and asking NVIDIA’s Nemotron-based NIM to produce clear, structured summaries that render live in a React dashboard, with optional on-device OpenCV/MediaPipe fall detection for privacy.
Stack
Process
Problem Framing
Researched elder-care challenges and focused on the gap between emergency-response devices and true day-to-day monitoring.
System Design
Designed a multi-sensor home monitoring architecture that could capture routine activity without being overly invasive.
Agentic Layer
Built an agent-based intelligence layer to interpret sensor activity+ generate meaningful updates from patterns.
User Experience
Created a user-facing interface for caregivers to view activity and alerts in a way that feels clear and actionable.
Results & Impact
- Won Grand Prize at NVIDIA x AWS Agentic AI Hackathon out of 1900+ participants
- Brought home $7k GPU, swag, and mentorship!'
- Active development on Nomi - shipping incrementally