DEV Community

Rushil Patel
Rushil Patel

Posted on AI-assisted

Local AI for Local Climates: Building an Offline Garden Planner with Gemma 3 & Ollama

I live in Vadodara, and most planting advice online assumes a climate that isn't mine: frost dates, spring and fall. Here the seasons are monsoon, winter and summer, and October to February is dry and mild. So I built a small app that gives planting advice for Indian cities, using an open model that runs entirely on my own computer.

What it does

You type your city, click What should I plant?, and get a short list of what to plant this week. The app picks the season (monsoon, winter or summer) from today's date, then asks Gemma (gemma3:4b), running locally through Ollama, to write the advice.

Code: https://github.com/RUSHILPATEL33/garden-planner

Why open-source mattered

  • Free. No API key and no per-question cost. After the one-time download, it costs nothing to run.
  • Offline. Once the model is downloaded, it works without internet, which suits a garden.
  • Private. My city never leaves my computer.

What went wrong first

My first answer had three problems: raw ** symbols showing on the page, advice about winter rain and frost that doesn't fit Gujarat, and the same vegetable listed twice under two names (radish and mooli).

I fixed it by telling the model that India's October to February is dry and mild, limiting it to 5 plants with no repeats, and rendering bold text and lists properly on the page. With a small local model, the prompt does most of the work.

What I'd do next

[One or two ideas, for example more cities, a care schedule for the chosen crop, or Hindi and Gujarati support.]

How I built it

I used Cursor's agent for the first version and DevRelay to connect to the Hacktoberfest challenges.

Top comments (0)