DEV Community

Yogesh Kumar
Yogesh Kumar

Posted on

I Built G-Buudddyy : Bringing AI-Powered Experiences Beyond the Screen

#mlhacks #devchallenge #hackathon #devcyclechallenge

AI Garden Buddy

Less scrolling. More growing.

AI Garden Buddy is a local-first gardening assistant that uses AI to help people start gardening. It recommends suitable plants, generates a personalized seven-day gardening plan, and suggests one small outdoor task to complete each day.

Built for the Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass.

Live Demo

Live Application: https://hacktoberfest-tawny.vercel.app/

GitHub Repository: https://github.com/yogesh968/hacktoberfest

Note: AI plan generation requires the FastAPI backend and Ollama. The deployed frontend needs a reachable backend to generate plans for online users.

Features

  • Personalized Plant Recommendations: Suggests plants based on city, garden type, sunlight, and gardening experience.
  • Seven-Day Gardening Plan: Generates a practical daily schedule based on available time.
  • Daily Outdoor Tasks: Provides actionable gardening activities with instructions and estimated completion times.
  • Progress Tracking: Allows users to mark tasks as complete and record reflections in their browser.
  • Safety Notes: Highlights considerations such as plant toxicity, heat exposure, heavy pots, and tools.
  • Uncertainty Notes: Explains limitations, including the lack of live weather and frost-date information.
  • Local AI Inference: Runs a Gemma model locally through Ollama.
  • Structured Output Validation: Uses JSON Schema and Pydantic to validate AI-generated plans.
  • Reliable Error Handling: Reports model failures instead of presenting fabricated AI-generated plans.

Tech Stack

Frontend

  • React 18
  • Vite 6
  • Tailwind CSS v4
  • lucide-react
  • Fontsource

Backend

  • Python
  • FastAPI
  • Pydantic v2
  • Uvicorn
  • httpx
  • pydantic-settings

AI and Testing

  • Google Gemma
  • Ollama
  • JSON Schema
  • pytest
  • ESLint

How It Works

  1. The user enters their city, garden type, sunlight conditions, experience level, and available daily time.
  2. The React frontend sends the information to the FastAPI backend.
  3. The backend sends a prompt and JSON Schema to a Gemma model through Ollama.
  4. Pydantic validates the generated response.
  5. The frontend displays the gardening plan, plant recommendations, safety notes, and daily task.
  6. The user completes the outdoor task, marks it as done, and records a reflection.

Architecture 

React + Vite Frontend
        |
        v
FastAPI Backend
        |
        v
Ollama Inference Runtime
        |
        v
Gemma AI Model
        |
        v
Validated Gardening Plan
        |
        v
Outdoor Task and Progress Tracking
Enter fullscreen mode Exit fullscreen mode

Getting Started

Prerequisites

  • Python 3.10 or later
  • Node.js and npm
  • Ollama
  • Sufficient memory and disk space for your selected AI model

1. Clone the Repository 

git clone https://github.com/yogesh968/hacktoberfest.git
cd hacktoberfest
Enter fullscreen mode Exit fullscreen mode

2. Download the AI Model

Install Ollama and run:

ollama pull gemma3:4b
Enter fullscreen mode Exit fullscreen mode

Alternatively, use the smaller model:

ollama pull gemma3:1b
Enter fullscreen mode Exit fullscreen mode

3. Set Up the Backend 

cd backend
python -m venv .venv
Enter fullscreen mode Exit fullscreen mode

Activate the environment.

Windows PowerShell:

.venv\Scripts\Activate.ps1
Enter fullscreen mode Exit fullscreen mode

macOS/Linux:

source .venv/bin/activate
Enter fullscreen mode Exit fullscreen mode

Install dependencies:

pip install -r requirements.txt
Enter fullscreen mode Exit fullscreen mode

Copy the example environment file:

# Windows
copy .env.example .env

# macOS/Linux
cp .env.example .env
Enter fullscreen mode Exit fullscreen mode

4. Set Up the Frontend

Open another terminal from the repository root:

cd frontend
npm install
Enter fullscreen mode Exit fullscreen mode

Copy the frontend environment file:

# Windows
copy .env.example .env

# macOS/Linux
cp .env.example .env
Enter fullscreen mode Exit fullscreen mode

For local development, configure the backend URL:

VITE_API_BASE_URL=http://localhost:8000
Enter fullscreen mode Exit fullscreen mode

5. Run the Application

Start Ollama if it is not already running:

ollama serve
Enter fullscreen mode Exit fullscreen mode

Start the backend from the backend directory:

uvicorn app.main:app --reload --port 8000
Enter fullscreen mode Exit fullscreen mode

Start the frontend from the frontend directory:

npm run dev
Enter fullscreen mode Exit fullscreen mode

Open the application:

Make sure the model configured in the backend has been downloaded.

API Overview

GET /api/health

Checks backend health and model connectivity.

POST /api/garden-plan

Generates a personalized gardening plan.

Example request:

curl -X POST http://localhost:8000/api/garden-plan \
  -H "Content-Type: application/json" \
  -d '{"city":"Pune","garden_type":"balcony","sunlight":"partial_sun","experience":"beginner","minutes_per_day":15}'
Enter fullscreen mode Exit fullscreen mode

Use the exact request fields and accepted values defined by the backend's Pydantic model.

Reliability and Error Handling

The backend uses multiple validation layers to improve reliability:

  • Ollama structured outputs constrain the response using JSON Schema.
  • Pydantic validates the generated response.
  • Additional checks enforce application-specific requirements, including seven days and between two and six plant recommendations.
  • Invalid output triggers one retry.
  • If validation still fails, the backend returns a clear error.

When AI generation is unavailable, the application can display general gardening advice labelled "Not AI-generated" instead of pretending it came from the model.

Testing

The backend includes 13 automated API tests using pytest.

The tests use a fake HTTP transport instead of Ollama, allowing API behavior and failure scenarios to be tested without running an actual model.

Coverage includes:

  • Valid plans
  • Malformed model output and retry behavior
  • Ollama being unavailable
  • Missing models
  • Request timeouts
  • Invalid form input

Run the backend tests:

python -m pytest -q
Enter fullscreen mode Exit fullscreen mode

Run frontend checks from the frontend directory:

npm run lint
npm run build
Enter fullscreen mode Exit fullscreen mode

Privacy

With the default local configuration:

  • User inputs are sent to the local FastAPI backend and Ollama service.
  • The backend does not log the city or other form inputs.
  • Task completion and reflections are stored in the browser.
  • Local AI inference does not require a hosted AI API key.

These privacy statements apply to the local configuration. A remote backend or cloud model may process data outside the user's computer.

Limitations

  • The application does not automatically retrieve live weather or frost-date information.
  • Plant recommendations may not account for every local growing condition.
  • Smaller AI models may occasionally generate invalid or imperfect responses.
  • Generation speed depends on the user's hardware.
  • Users should verify plant toxicity and local growing advice, especially around children and pets.

Model Licensing

Gemma models have their own licensing terms. Gemma 3 is distributed under Google's Gemma Terms of Use, which differ from an OSI-approved open-source license.

Check the applicable terms for the specific model version you use. The application's code license and the AI model's license are separate.

Contributing

Contributions and suggestions are welcome.

  1. Fork the repository.
  2. Create a branch for your changes.
  3. Implement your changes and add tests where appropriate.
  4. Run the relevant tests and code-quality checks.
  5. Submit a pull request describing your changes.

Please do not commit secrets, environment files, or other sensitive information.

Future Improvements

  • Integration with verified weather and climate information
  • Multilingual gardening assistance
  • Plant images with appropriate attribution
  • Saved gardening plans and streaks
  • Improved regional plant recommendations
  • Simplified local setup using Docker Compose

Acknowledgements

Built for the Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass.

AI Garden Buddy explores how locally run AI can help people move from browsing and learning to taking meaningful action in the real world.

Less scrolling. More growing.

Top comments (0)