This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Plant Care Companion — A free, AI-powered web app that gets people off their screens and into their gardens.

Instead of doomscrolling, users step outside, snap a photo of any plant (houseplant, garden flower, random leaf on a walk), and instantly get:

  • Plant identification with scientific names and fun facts
  • Health diagnosis detecting diseases, pests, overwatering, nutrient issues
  • 3-step action plans to fix problems in plain, beginner-friendly language

It's built for anyone who wants to care for plants but lacks botanical knowledge — busy parents, apartment dwellers with struggling succulents, curious kids exploring parks, or hobby gardeners troubleshooting brown leaves.

The app encourages real-world interaction: you have to physically go outside, observe your plant closely, and take action based on the diagnosis. The screen is just the bridge, not the destination.

Demo

🌐 Live App: https://plant-care-ai-u9u8.vercel.app/

📸 How it works:

  1. Upload a plant photo (drag & drop or click)
  2. Choose "Identify this Plant" or "Check Health & Diseases"
  3. Get instant, warm, human-like expert advice in seconds

![App Screenshot](

Code

GitHub logo CYBERtony786 / plant-care-ai

🌿 Plant Care Companion

Built for Hacktoberfest 2026 - Week 1: Touch Grass Challenge

A modern, AI-powered web application that helps people step away from their screens and connect with nature. Upload a photo of any plant, and get instant expert identification and health advice.

🌍 Problem Statement

In today's digital world, people spend countless hours staring at screens while their houseplants wither away or their gardens remain mysteries. Many want to care for plants but lack botanical knowledge. Commercial plant apps are expensive, require subscriptions, and often share your data.

Plant Care Companion solves this by:

  • Providing free, instant plant identification
  • Diagnosing plant health issues with actionable fixes
  • Encouraging people to go outside, observe nature, and engage with their plants
  • Keeping your data private and secure

✨ Features

🌱 Plant Identification

Upload a photo and instantly learn the common name, scientific name, and interesting facts about your plant.

🩺

…
View on GitHub

Key files:

  • app/page.tsx — Main UI with image upload and action buttons
  • app/api/analyze/route.ts — Serverless API that calls Gemini Vision AI
  • app/components/ImageUploader.tsx — Drag-and-drop component
  • .env.local — Secure API key storage (gitignored)

How I Built It

Open-Source AI at the Core

I used Google Gemini 3.8 Flash (vision-capable, open-weight model) via the @google/genai SDK. This model can analyze images and respond in natural language — perfect for identifying plants and diagnosing issues from photos.

Why Gemini?

  • Free tier with generous limits (no per-token costs during development)
  • Vision capabilities out of the box (processes photos directly)
  • Fast inference (~2-3 seconds per request)
  • Can be swapped for local models like LLaVA if I want to run this offline

Tech Stack

  • Frontend: Next.js 15 + React 19 + TypeScript
  • Styling: Tailwind CSS 4 with custom earthy green theme (not generic AI blue!)
  • Backend: Next.js API Routes (serverless functions)
  • Deployment: Vercel (free tier, auto-scaling)

Architecture

The app uses a simple client-server flow:

  1. User uploads image → converted to base64 in browser
  2. Frontend sends base64 + prompt to /api/analyze
  3. Backend calls Gemini Vision API with custom prompts
  4. AI response is formatted and returned to user

I crafted two distinct prompts:

  • Botanist mode: "You are a friendly expert botanist..." (warm, educational)
  • Gardener mode: "You are an experienced gardener..." (practical, troubleshooting)

This makes the AI sound human, not robotic.

Why Does Open Innovation Matter?

Privacy & Control

Using open-weight models means users could run this locally (with Ollama + LLaVA) without sending plant photos to third-party servers. For rare plant collectors or commercial growers, keeping image data private matters.

Cost & Accessibility

Closed APIs like GPT-4 Vision charge per-token. For a free gardening tool, that's unsustainable. Open models (Gemini free tier, or self-hosted Llama Vision) make this viable long-term.

Transparency & Customization

I can see exactly how the model works, fine-tune prompts, and even swap models if Google's API changes. With a closed API, I'm locked in. Open AI lets me adapt — for example, I could train a specialized model on plant diseases specific to my region.

Offline Access

In the future, I could package this as a Progressive Web App (PWA) with a local model for hikers and nature explorers with no internet connection. Open AI makes that possible.

My Agent Session

This entire project was built using "Vibe Coding" with an AI agent (Antigravity). I wrote zero lines of code by hand — just described what I wanted in plain language, and the agent:

  • Scaffolded the full Next.js project structure
  • Wrote TypeScript components with drag-and-drop file upload
  • Integrated Gemini Vision API with error handling
  • Created a professional README for GitHub
  • Fixed API versioning issues when Google updated model names

(Optional: If you saved your session with DevRelay, embed it here using the agent_session tag)

Prize Categories

I'm entering:

  • Best Overall Project (Core challenge)
  • Best Use of Open-Source AI (Uses open-weight Gemini model)
  • Most Impactful "Touch Grass" Project (Gets people outdoors and gardening)

Built with ❤️ and open-source AI for Hacktoberfest 2026.

Get outside. Touch grass. Care for your plants. 🌱