AI Gardener

A smart, offline-first autonomous agent that monitors and cares for plants using Edge AI and RAG.

Overview

AI Gardener moves beyond simple 'if-moisture-low' logic. By leveraging Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG), the system understands the specific needs of different plant species. Whether it's a drought-resistant cactus or a water-loving pothos, the AI Gardener retrieves the correct care parameters and executes precise watering cycles without needing an internet connection.

Gallery

View

Tech Stack

Jetson Nano,

Ollama (Qwen2.5/TinyLlama)

RAG (Sentence-Transformers)

Soil Moisture Sensors

ESP32-S3

MQTT (Mosquitto)

Laravel

Python (FastAPI)

Node.js

Communication: JSON-based Webhooks, Pub/Sub Messaging

Key Features

Species-Aware Intelligence: Uses RAG to fetch specific care instructions for different plants before making a decision.

Edge-Native Autonomy: Runs Ollama and localized LLMs on Jetson Nano, ensuring your plants are cared for even if the internet goes out.

Asynchronous Cloud Sync: Triggers webhooks to a Laravel/Node.js backend for long-term health tracking and remote monitoring.

Low-Power IoT Nodes: Efficient ESP32-S3 clusters that manage multiple plant sensors and actuators.

Real-time Analytics: A professional-grade dashboard for visualizing soil health, temperature trends, and AI action logs.

Challenges & Solutions

Ensuring the AI agent remained focused on hardware commands was a key challenge. I designed a custom 'Instruction-Response' pipeline that constrains the LLM to output only executable commands. Managing the hybrid network state (local MQTT vs. remote Webhooks) also required a robust error-handling logic in Python to prevent data loss.

Outcome & Impact

The result is a fully functional AI Gardener that mimics human judgment in plant care. The system effectively manages multiple plant nodes, reduces water waste, and provides a seamless bridge between local hardware automation and cloud data visualization.

Explore AI Gardener