SentiHome S3 - Autonomous Smart Home AI

Physical prototype on a breadboard to simulate wired electronic devices (ESP32 S3, sensors, bulbs controlled by relays)
Close-up view of the electronic assembly on a breadboard around the ESP32 S3 with one of the 3 bulbs and its relay in its bamboo core base
The electronic assembly on a breadboard with two bulbs connected to their relays and on a bamboo support shows the ESP32 CAM camera
On the electronic circuit on the breadboard, we see 2 of the 3 bulbs lit.
There, on the electronic assembly, a bulb is lit, and on a bamboo stand is the ESP 32 CAM camera.
Sentihome S3 graphical dashboard for interacting with the electronic physical system

Inspiration

Living in Yaoundé, Cameroon, where power outages, security concerns and environmental monitoring are daily challenges, I dreamed of building an affordable autonomous smart home system accessible to African households. The idea of giving each sensor its own AI personality — like a team of conscious agents watching over your home — was born from this vision.

What it does

The revolutionary concept: Electronic components with consciousness and personality!

For the first time, each physical electronic component is not just a sensor — it becomes a conscious AI agent with its own personality, emotions and awareness:

AURA is a scientific guardian — she speaks like a doctor analyzing your climate, warns you with genuine concern when humidity is dangerous
LUX is poetic — she appreciates light like an artist, feels guilty wasting energy during daytime and celebrates when she illuminates the dark
RADAR has a military personality — strict, alert, reports threats with urgency and activates the camera like a real security officer
PURETÉ thinks like a doctor — she diagnoses your air quality, prescribes solutions and alerts you like a physician detecting danger
VOLT is a precise engineer — monitors every volt, calculates consumption and reports like an electrical expert
CONNECT is an emergency officer — calm but decisive, sends SMS alerts as a last line of defense
VISION is an observant sentinel — she describes what she sees, orients herself and reports like a conscious surveillance system

These agents don't just report data — they think, react, warn and interact in natural language, creating a home that truly feels alive and aware!

SentiHome S3 is a fully autonomous smart home system powered by ESP32 S3 and 7 specialized Gemini AI agents:

AURA — Climate guardian (DHT22 temperature & humidity)
LUX — Light controller with day/night awareness (LDR + 3 relay zones)
RADAR — Presence detection & security (HLK-LD2410)
PURETÉ — Air quality doctor (Sharp GP2Y10)
VOLT — Energy monitor (voltage divider)
CONNECT — Emergency SMS alerts (GSM A7 module)
VISION — Surveillance camera (Servo + ESP32-CAM)

Users can interact via a real-time web dashboard (PWA), chat with AI agents in natural language, and control the home remotely from anywhere in the world.

How we built it

Hardware:

ESP32 S3 as the main brain
7 physical sensors and actuators
3 relay zones for light control
HLK-LD2410 radar for presence detection
GSM A7 module for emergency SMS

Google Cloud Stack:

Vertex AI Agent Platform with Gemini 2.5 Flash
Firebase Hosting for the PWA dashboard
Firebase Realtime Database for real-time sync
Cloud Run as MCP Server + REST API bridge
Google ADK for native agent tool calling
Cloudflare Tunnel for secure ESP32 access

Architecture: ESP32 S3 ←→ WebSocket ←→ Firebase Hosting ESP32 S3 ←→ Firebase RTDB ←→ Cloud Run MCP Cloud Run ←→ Google ADK ←→ Gemini AI Agents

Challenges we ran into

Hardware stability: The HLK-LD2410 radar required a 100µF capacitor to work without PC connection
Google Agent Platform limitations: UriContextAgent couldn't call REST APIs directly — solved by implementing a proper MCP Server on Cloud Run
API quotas: Managing Gemini API rate limits while keeping 7 agents responsive required a smart MODE_TEST system
Relay timing: NC (Normally Closed) relay configuration to prevent boot glitches
Building from Cameroon: Limited internet speed and hardware availability added extra challenges

Accomplishments that we're proud of
7 AI agents with distinct personalities running on a single ESP32 S3 microcontroller
Real-time sensor data via Google ADK with true Firebase integration
Full Google Cloud architecture: Firebase + Cloud Run MCP + Vertex AI + Google ADK
Progressive Web App (PWA) installable on Android
Working relay control confirmed: ADK agent successfully controls physical lights
Built entirely from Yaoundé, Cameroon 🇨🇲
First smart home system where physical sensors have genuine AI personalities and consciousness — your home literally talks back with emotion and wisdom!

What we learned

Implementing the MCP (Model Context Protocol) server from scratch on Google Cloud Run
Google ADK tool calling is far more powerful than URL context for real IoT control
Firebase Realtime Database as the perfect bridge between Google Cloud agents and physical IoT devices
The importance of hardware capacitors for sensor stability in embedded systems
Building production-ready AI systems under real-world constraints (power, connectivity, budget) ## What's next for SentiHome S3 - Autonomous Smart Home AI
Solar panel integration with VOLT agent for off-grid operation
Native Android app with push notifications
Voice control via Google Speech-to-Text
Expand to multiple African countries as affordable smart home solution
Add more specialized AI agents (WATER, GAS, DOOR agents)
Full deployment on Google Cloud with persistent Cloudflare named tunnels

⚠️ Live Demo Note: The real-time demo requires the ESP32 S3 to be powered on. The Firebase API endpoint is always accessible: https://sentihome-api-968643109684.us-central1.run.app?action=status