惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

罗磊的独立博客
T
Tenable Blog
人人都是产品经理
人人都是产品经理
IT之家
IT之家
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
小众软件
小众软件
美团技术团队
The GitHub Blog
The GitHub Blog
Y
Y Combinator Blog
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
V
Visual Studio Blog
M
Microsoft Research Blog - Microsoft Research
aimingoo的专栏
aimingoo的专栏
P
Proofpoint News Feed
T
The Blog of Author Tim Ferriss
博客园 - 聂微东
V
V2EX
Microsoft Security Blog
Microsoft Security Blog
C
CXSECURITY Database RSS Feed - CXSecurity.com
爱范儿
爱范儿
Latest news
Latest news
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
I
InfoQ
H
Help Net Security
Google DeepMind News
Google DeepMind News
P
Privacy International News Feed
U
Unit 42
Cyberwarzone
Cyberwarzone
V
Vulnerabilities – Threatpost
F
Future of Privacy Forum
雷峰网
雷峰网
Recorded Future
Recorded Future
WordPress大学
WordPress大学
P
Privacy & Cybersecurity Law Blog
博客园 - Franky
D
Darknet – Hacking Tools, Hacker News & Cyber Security
N
Netflix TechBlog - Medium
D
Docker
博客园_首页
J
Java Code Geeks
CTFtime.org: upcoming CTF events
CTFtime.org: upcoming CTF events
Blog — PlanetScale
Blog — PlanetScale
C
CERT Recently Published Vulnerability Notes
Malwarebytes
Malwarebytes
MongoDB | Blog
MongoDB | Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Cisco Talos Blog
Cisco Talos Blog
T
Threat Research - Cisco Blogs
Know Your Adversary
Know Your Adversary
GbyAI
GbyAI

DEV Community

Git Time Machine — How Version Control Can Save Your Project My Dad Got an Electricity Bill He Couldn't Understand. Google I/O 2026 Just Made That Problem Solvable. My Dad Got an Electricity Bill He Couldn't Understand. Google I/O 2026 Just Made That Problem Solvable. Read Replicas Lie About Consistency. 4 Sync Modes Behind the Lie. Reviving My Coding Project with GitHub Copilot I Tried Gemini 3.5 Flash After Google I/O 2026 - Here is What I Found :)) Zero-Cost AI in VS Code Blueprints Might Be More Important Than Frameworks Long-Context Models Killed RAG. Except for the 6 Cases Where They Made It Worse. I Built a Neural Network Engine in C# That Runs in Your Browser - No ONNX Runtime, No JavaScript Bridge, No Native Binaries An In-Depth Overview of the Apache Iceberg 1.11.0 Release Your Agent Just Called the Same Tool 47 Times. Here's the 20-Line Detector. How I Built a Multi-System Astrology Bot in Python (And What Meta Banned Me For) Gemma 4 Has Four Variants. Here's How to Pick the Right One Before You Write a Single Line of Code. Log Level Strategies: Balancing Observability and Cost Why WebMCP Is the Most Important Thing Google Announced at I/O 2026 (And Nobody's Talking About It) Making LLM Calls Reliable: Retry, Semaphore, Cache, and Batch Google's 2x Energy Efficiency Claim Is Real — But Here's What They're Not Measuring What's actually going on with CORS, under the hood Language-Agnostic Code Generation: The Driver Plugin Model Why We Rewrote Our Python CLI in Go (and What We Gained) I added up everything Google gives developers for free after I/O 2026. It's kind of absurd The Dawn of Smarter Apps: My Take on Google I/O 2026 AI Announcements Why AI Agents Like Hermes Need a Semantic Execution Layer for the Physical World Why We Built TestSmith: The Test Coverage Problem Nobody Talks About How to Convert Bank Statement PDFs to Excel: The Complete 2026 Guide Have You Ever Used a Website That Keeps Working After You Turn Off Your Internet? From idea to indexed: how I launched a SaaS in 60 days with Laravel + React Building a local-first AI tutor for my daughter (and 10–14 year-olds in Austrian schools) with Gemma 4 EC2 SSH Not Connecting? Here Are the 5 Things That Were Wrong (And How I Fixed Them) Best AI Tools for HVAC Contractors 2026 From Closed Internal Stack to Open-Source Ecosystem: I Finally Shipped Three Years of .NET Infrastructure Scrumpan is offlically LIVE!! Building a BMI Calculator CLI with TypeScript — Types, Functions, and Vitest From Building WordPress Websites to Node.js APIs: My Honest Full Stack Journey XiHan Snore Coach: Privacy-First On-Device MedTech Guardian powered by Gemma 4 Mobile Why AI Coding Agents Hallucinate and How to Fix It mcp-probe v1.4.0: Contract assertions for production MCP servers Google I/O 2026 Wasn't About One More Model. It Was About the Agent Stack. How I built 100+ crypto calculators in 6 languages on Astro The Dawn of Local Multi-Agent Architectures: Why Gemma 4 Changes Everything for Cloud Developers # I Told My AI to Simulate a Planet for 10,000 Years. It Built the Whole Thing Itself. 18/30 Days System Design Questions! From Hackathon Chaos to Clean CLI: Reviving My Daily Routine Analyser with GitHub Copilot Building a Home Lab with Proxmox and Terraform (for Kubernetes) PolicyAware vs Guardrails vs AI Gateways vs Model Routers: The Comparison Every AI Engineer Needs to Read Partner: An AI That Does Research While You Sleep Rugby Fundamentals as Software Concepts - Mapping the Pitch to your Code Base I Let Claude Code Run Unsupervised for 24 Hours. Here's What Happened. Why Zed Is Replacing VS Code in My AI-Augmented Workflow Build a scroll-driven WebGL hero in 30 lines Karpathy's LLM Wiki? No Code with Claude or Github Copilot! Why Platform Governance and Transparency Matter for Developers and Freelancers I built a Flutter CLI that generates Clean Architecture in seconds Using an LLM to automate a task that used to take hours by hand CyberArena – Interactive Cyber Security Simulation & Threat Analysis Platform Tile Extractor Mathematical Functions in CSS: clamp, min, max and How They Simplify Responsiveness Polyglot Persistence in Microservices: Let the Domain Choose the Database 190 Countries, Zero API Calls: Shipping Static Data in a Chrome Extension Your AI Writes Code Fast. Here’s How to Check It Before Shipping qwen2.5-coder is too slow for Claude Code on a Mac. Here's the fix. Building Automated Text-to-Video Pipelines with AI Can Gemini Become an Offline AI Tutor? Lessons from Building Educational AI OPRIX : From a simple messaging web app to a well structured and enhanced UI messaging web app Why React + TypeScript Nullability Slowly Becomes Exhausting Why AI Agents Need a Project Layer - Part 1 Stop Hand-Editing MCP Configs: A Zero-Dependency Go CLI What I Learned Working With Microsoft, SQUAD(GTCO), and Different Tech Communities 🧠 Hermes Agent Assistant — A Modular AI Agent System with Planner, Executor & Memory Spring Boot Auto-Configuration Source Code: Nail This Interview Question The Ultimate Guide to Free AI API Keys: 6 Platforms You Need to Know Why 91% of AI Agents Fail in Production (And What the 9% Do Differently) TryHackMe | Battery | WALKTHROUGH Stop Guessing Your Regex — Test It Live in the Browser I Built FreelancEye, an Open-Source Mobile PWA for Finding Clients Beyond the Hype: My Production Playbook for Docker Swarm Top AI App Builder Platforms with Integrated Backend, Hosting & Database ECS vs EKS in 2026: An Honest Comparison from Someone Who Has Run Both in Production Hardening Your Node.js App Against Supply Chain & Remote Code Execution Attacks linux commands A Practical GEO Case: How an AI System Started Recommending Our Blog Your AI Agent Works 24/7 and Earns $0. I Built the Fix. Your AI Trading Agent Will Lose All Your Money — Here's How To Stop It Google I/O 2026: What Happens When Everything Connects? Why AI writes software but doesn’t build a good product Beyond the Hype: How Google I/O 2026 Secretly Democratized Production-Ready AI Agents with Managed Sandboxes. The Killer Assumption Test: How to Spot Doomed Product Decisions Before You Ship Stop Describing Your Bugs — Just Screenshot Them # I Built an AI Website Builder and Here's What Actually Happened Cooking an AI Campaign in 5 Minutes with Google Cloud AI APIs Your PM Retrospectives Are Lying to You How I Built a Free, Self-Hosted Pipeline That Auto-Generates Faceless YouTube Shorts TypeScript 54 to 58: The Features That Actually Matter in 2026 How to Tailor Your CV to Any Job Posting in 2026 The 7-day SaaS MVP loop: ship fast, then validate with people who actually show up 95. Fine-Tuning LLMs: Make a General Model Do Your Specific Job What Is a Frontend Developer Roadmap and Why You Need One Google shipped three Gemini "Flash" models. Picking the wrong one could 6 your AI bill Building an MCP server so Claude can query my SaaS analytics directly
AI CareCompanion - Offline Health Assistant
NARENDER KES · 2026-05-24 · via DEV Community

NARENDER KESWANI

Gemma 4 Challenge: Build With Gemma 4 Submission

Offline-first AI health companion powered by Google's Gemma 4 4B running entirely on-device. Multi-agent specialist routing, drug lookup, symptom triage, nearby doctor search, and Firebase-backed cloud sync — all in a single React Native app for Android and iOS.


YouTube Video

AI CareCompanion Demo


Screenshots

Splash Screen

Splash Screen


Sign In

Sign In


Model Download

Model Download


Home Dashboard

Home


Lab Report Analysis

Lab Report Analysis


AI Response

AI Response


Skin Lesion Analysis

Skin Lesion Analysis


Hindi Language Support

Hindi Language


Chat Reports

Chat Reports


Generated Report

Generated Report


PDF Patient Summary

PDF Patient Summary


PDF Clinical Report

PDF Clinical Report


Health Insights

Health Insights


Insight Detail

Insight Detail


Navigation Menu

Navigation Menu


Drug Search

Drug Search


Drug Detail

Drug Detail


Affordable Alternatives

Affordable Alternatives


Nearby Doctors

Find Nearby


On-Device AI

On-Device AI


System Architecture

High-Level Design (HLD)

High Level Design

Data flow between the Mobile Client, AI Orchestrator, Inference Engines, and SQLite/Firebase Data Layer.


Low-Level Design (LLD)

Low Level Design

The 5-step sequence flow of a user query being routed and processed by the AI Orchestrator and Specialists.

Features

On-Device AI — No Internet Required

  • Gemma 4 4B (IQ2_M) runs fully on-device via llama.rn (GGUF format)
  • ~1.5 GB model, ~2 GB RAM — works offline after one-time download
  • Zero data sent to external servers during inference
  • Cloud AI (Gemini) used as fallback when online

Multi-Agent Specialist System

  • 46 specialist agents — each with domain-specific system prompts and tool access
  • Central orchestrator auto-detects intent and routes to the right specialist(s)
  • Up to 3 agents collaborate on complex multi-system queries
  • Real-time routing info (agent name + response time) shown in every message
# Agent Specialty Image Support
1 Triage Assistant Orchestrator — intent detection & routing
2 General Health Specialist General Practice
3 Child Health Specialist Paediatrics
4 Chronic Care Specialist Internal Medicine
5 Heart Health Specialist Cardiology
6 Skin Care Specialist Dermatology
7 Brain & Nerve Specialist Neurology
8 Digestive Health Specialist Gastroenterology
9 Lung & Breathing Specialist Pulmonology
10 Kidney Specialist Nephrology
11 Hormone Specialist Endocrinology
12 Infection Specialist Infectious Disease
13 Women's Health Specialist Obstetrics & Gynecology
14 Bone & Muscle Specialist Orthopaedics
15 Eye Care Specialist Ophthalmology
16 Ear, Nose & Throat Specialist ENT
17 Mental Health Specialist Psychiatry
18 Dental Care Specialist Dentistry
19 Urinary Tract Specialist Urology
20 Blood Specialist Haematology
21 Joint & Immune Specialist Rheumatology
22 Allergy Specialist Allergy & Immunology
23 Newborn Care Specialist Neonatology
24 Senior Health Specialist Geriatrics
25 Emergency Care Specialist Emergency Medicine
26 Nutrition & Diet Specialist Nutrition & Dietetics
27 Medication Specialist Pharmacy
28 Radiology & Imaging Specialist Radiology
29 Sports Medicine Specialist Sports Medicine
30 Sleep Medicine Specialist Sleep Medicine
31 Palliative & Hospice Care Specialist Palliative Care
32 Rehabilitation Specialist Physiatry / Rehab Medicine
33 Genetics & Genomics Specialist Genetics
34 Pain Management Specialist Pain Management
35 Toxicology & Poisoning Specialist Toxicology
36 Travel Medicine Specialist Travel Medicine
37 Adolescent Medicine Specialist Adolescent Medicine
38 Reproductive & Fertility Specialist Fertility
39 Vascular Medicine Specialist Vascular Medicine
40 Transplant Medicine Specialist Transplant Medicine
41 Clinical Pathology Specialist Pathology
42 Integrative Medicine Specialist Integrative Medicine
43 Addiction Medicine Specialist Addiction Medicine
44 Occupational Medicine Specialist Occupational Medicine
45 Lifestyle Medicine Specialist Lifestyle Medicine
46 Preventive Medicine Specialist Preventive Medicine

Health Tools

  • Drug Info — Search 1mg medicine database (name, dosage, warnings, interactions)
  • Affordable Alternatives — Find cheaper generic substitutes with price comparison
  • First Aid Guide — Step-by-step emergency guidance
  • Find Nearby Doctors — Location-based doctor and hospital search
  • Image Analysis — Analyze wound, rash, or medical images via camera/gallery
  • Health Metrics — Track vitals (BP, glucose, weight) over time
  • Medication Reminders — Local notifications for medication schedules
  • Chat Reports — Export conversation summaries as PDF

Privacy & Sync

  • All AI inference runs on-device — no health data leaves your phone
  • SQLite local database for all chats, profiles, and history
  • Firebase Firestore optional cloud sync with incremental pull + outbox pattern
  • Biometric authentication (fingerprint/face) for app lock

UI/UX

  • Light and Dark mode
  • Markdown rendering in AI responses
  • Animated splash screen
  • Search history across drug lookups and AI queries

Tech Stack

Layer Technology
Framework React Native 0.76 (TypeScript)
On-Device AI llama.rn — Gemma 4 4B IQ2_M GGUF
Cloud AI Google Gemini (online fallback)
Auth Firebase Auth (Email/Password + email verification)
Local DB SQLite via react-native-sqlite-storage
Cloud Sync Firebase Firestore
Navigation React Navigation v7
Biometrics react-native-biometrics

Setup

Prerequisites

  • Node.js >= 18
  • React Native CLI
  • Android Studio (API 33+) or Xcode 15+
  • Firebase project with Auth and Firestore enabled

1. Clone and install

git clone https://github.com/narender-rk10/AI-CareCompanion-Offline-Health-By-Gemma.git
cd AI-CareCompanion-Offline-Health-By-Gemma
npm install

Enter fullscreen mode Exit fullscreen mode

2. iOS pods

cd ios && pod install && cd ..

Enter fullscreen mode Exit fullscreen mode

3. Firebase configuration

  1. Create a project at console.firebase.google.com
  2. Enable Authentication → Email/Password
  3. Enable Firestore Database
  4. Download google-services.json → place in android/app/
  5. Download GoogleService-Info.plist → add to Xcode project

4. Environment variables

cp .env.example .env

Enter fullscreen mode Exit fullscreen mode

Then fill in .env:

# Gemini API key — https://aistudio.google.com/app/apikey
GEMINI_API_KEY=your_gemini_api_key_here

# Google Sign-In Web Client ID (Firebase Console → Auth → Google → Web SDK config)
GOOGLE_WEB_CLIENT_ID=your_google_web_client_id.apps.googleusercontent.com

# Direct GGUF download URL for Gemma 4 4B (~1.5 GB)
MODEL_DOWNLOAD_URL=https://huggingface.co/your-org/model/resolve/main/model.gguf

Enter fullscreen mode Exit fullscreen mode

Firebase credentials go in android/app/google-services.json and ios/GoogleService-Info.plist — not in .env.

5. Run

# Android
npm run android

# iOS
npm run ios

Enter fullscreen mode Exit fullscreen mode

6. Download the AI model

On first launch, the app will prompt you to download Gemma 4 4B (~1.5 GB). Requires a stable Wi-Fi connection and ~3 GB free storage. After download the model runs fully offline.

Project Structure

src/
├── components/          # Reusable UI components
│   ├── ui/              # AppButton, AppInput, AppCard, ScreenHeader
│   ├── ChatBubble.tsx
│   ├── BiometricGate.tsx
│   ├── MarkdownText.tsx
│   └── SyncStatusBar.tsx
├── screens/             # All app screens
│   ├── ChatScreen.tsx               # Main AI chat
│   ├── HomeScreen.tsx               # Dashboard
│   ├── DrugInfoScreen.tsx           # Medicine search
│   ├── AffordableAlternativesScreen.tsx
│   ├── FirstAidGuideScreen.tsx
│   ├── FindNearbyDoctorScreen.tsx
│   ├── ImageAnalysisScreen.tsx
│   ├── AgentSelectorScreen.tsx
│   ├── ChatReportsScreen.tsx
│   ├── SettingsScreen.tsx
│   └── ...
├── services/            # Business logic
│   ├── LlmService.ts              # Gemini cloud AI
│   ├── LocalLlmService.ts         # llama.rn on-device AI
│   ├── AgentService.ts            # Multi-agent orchestration
│   ├── ModelManager.ts            # Model download & lifecycle
│   ├── DatabaseService.ts         # SQLite wrapper
│   ├── FirebaseService.ts         # Auth + Firestore sync
│   ├── DrugInfoService.ts         # 1mg medicine DB
│   └── PlaceSearchService.ts      # Nearby doctor search
├── context/
│   └── HealthProfileContext.tsx   # Global health profile state
├── navigation/
│   └── AppNavigator.tsx           # Auth flow + tab navigation
└── theme/                         # Colors, typography, dark mode

Enter fullscreen mode Exit fullscreen mode


Demo Prompts

"I have chest pain and shortness of breath"
"What is the dosage for metformin 500mg?"
"My child has had a fever for 3 days"
"Find a cardiologist near me"
"What are affordable alternatives to Crocin?"
"Analyze this rash on my arm" (attach photo)

Enter fullscreen mode Exit fullscreen mode


Privacy

  • AI inference is 100% on-device when using the offline model
  • Health profile and chat history stored only in local SQLite by default
  • Firebase sync is opt-in — no data uploaded without user action
  • Biometric lock prevents unauthorized access

License

Copyright (c) 2026 Narender Keswani. All Rights Reserved.

This project is proprietary software. You may view the source code for personal reference, but you may not copy, distribute, modify, or use it commercially without explicit written permission.

See LICENSE for full terms.


Built for people who need reliable health guidance anywhere — with or without internet.