This is a submission for the Gemma 4 Challenge: Build with Gemma 4
Your family generates health data every single day. Steps, sleep, heart rate, blood oxygen, passively, silently, constantly. But when something happens, no one had any warning. As a student doctor, I keep seeing the same thing: families with phones full of health data, and no one connecting the dots until they're already in a consultation room.
What I Built
HealthOps is a private health intelligence system for families living under one roof.
Your Android phone is already a capable health IoT device. Through Android HealthConnect, it continuously aggregates data from every health and fitness app installed, steps, sleep stages, resting heart rate, blood oxygen, and more, into a single local API. HealthOps taps into that, polling each family member's phone every 15 minutes, syncing delta records to a home server over local wifi, and building a time-series health picture per person. Gemma 4 then turns those raw signals into structured observations and plain-language narratives the whole family can actually use.
Everything stays in the house and you can get started in 10 minutes.
Demo
GitHub Releases — Download APK
Docker Image
Code
Testing Instructions
On the server (any computer on your home WiFi):
Make sure Docker is installed, then run the script below for your OS. It will start everything automatically.
Mac / Linux:
curl -fsSL https://raw.githubusercontent.com/CijeTheCreator/healthops/main/run.sh -o /tmp/run.sh && bash /tmp/run.sh
Windows (PowerShell):
irm https://raw.githubusercontent.com/CijeTheCreator/healthops/main/run.ps1 -OutFile "$env:TEMP\run.ps1"; powershell -ExecutionPolicy Bypass -File "$env:TEMP\run.ps1"
On startup, you'll be asked whether you want fully private local mode or whether you have a Gemini API key. Once running, the server address you'll need will be printed in the terminal. The app will be accessible from any device on your network.
On each Android phone:
Install the APK. Open it and fill in three things: your name, your WiFi network name, and the server address printed in the terminal. That's it, the phone will start syncing automatically.
Repeat the phone setup for each family member.
How I Used Gemma 4
HealthOps runs entirely on Gemma 4, locally, on the home server, and the pipeline has three stages:
1. Signal processing
Raw HealthConnect records arrive as delta pushes from each phone. Gemma converts them into structured observations of the form {signal, observation, time}, interpreting patterns across a rolling window of the last 30 health points rather than treating each record in isolation.
2. The Q&A agent
Any family member can ask natural language questions about the health data within a given time window: "How has Dad's resting heart rate trended this month?" or "Did anyone sleep poorly this week?" Gemma reasons across the full time-series for any member.
3. The weekly digest
Every week, a report agent assembles a plain-language health narrative for each family member.
Why Gemma 4 specifically
HealthOps uses two model configurations depending on how you set it up.
In fully private mode, it runs Gemma 4 E4B. It runs on basically any laptop made in the last 3-4 years, fast enough for real-time responses, and light enough to run quietly alongside your other apps. For most families, this is all they need.
If you opt in with a Gemma API key, HealthOps steps up to Gemma 4 27B A4B. Since we're no longer bound by local hardware constraints, we can afford a significantly larger model, which shows in the quality of the weekly digest and the Q&A agent.
Hardware
M1 Pro MacBook Pro (Server)
Google Pixel 10 Pro XL(Emulator)