My Garden World

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

What I Built

My Garden World — Discover nature. Grow with intelligence. Watch life unfold.

My Garden World is an AI-powered nature companion designed to encourage people to step away from their screens and reconnect with the natural living world.

It brings together three core experiences:

• AI-Powered Plant Discovery: Identify real-world flora on-device and explore botanical facts, sunlight, watering needs, and ecological importance.
• AI-Powered Garden Care & Vital Diagnostics: Get species-tailored treatment plans, NASA air-filtering scores, ASPCA pet-safety advisories, and photo growth tracking.
• Virtual Garden Builder & Nature Quests: Build and customize an interactive 8×8 sanctuary grid, complete outdoor mindfulness quests, and earn Flora Coins and XP.

The goal is to make learning about plants and gardening interactive, accessible, and enjoyable for beginners and nature enthusiasts alike.

Demo

Live Demo: https://my-garden-world-tau.vercel.app/

Explore the application and discover how technology can make outdoor exploration more engaging.

Code

GitHub Repository: https://github.com/jaya030888/my-garden-world

How I Built It

I built My Garden World as a modern, local-first full-stack application:

• Frontend: React 19, TypeScript, Vite, and Tailwind CSS deployed on Vercel.
• On-Device Vision AI: Client-side Vision Transformer (google/vit-base-patch16-224) executed locally via Transformers.js and ONNX Runtime Web (WebAssembly).
• Backend: Node.js and Express REST API with Helmet security and rate limiting.
• Database: Serverless PostgreSQL hosted on Neon, with Prisma ORM for schema migrations and relational data management.
• Local Persistence: Dexie.js (IndexedDB) caching for instant offline state resilience.

Why Does Open Innovation Matter?

For me, open innovation means having the freedom to build AI that serves people and nature—not being constrained by proprietary, black-box APIs.

With My Garden World, open-weight AI unlocks fundamental advantages:

1. Complete User Privacy Outdoors: Because the Vision Transformer executes directly in the user's browser sandbox via WebAssembly, no plant photographs ever leave the user's device.
2. Zero Proprietary Cloud Tax: We avoid expensive per-token closed-source vision APIs. The intelligence belongs to the user, running on their own phone or laptop at $0 inference cost.
3. Independence and Adaptability: Instead of being locked into a single provider's terms or pricing, the community can inspect, benchmark, and swap open-source vision architectures (ViT, MobileNet, Swin) freely.

The philosophy is simple: use open technology not to keep us glued to screens, but to inspire us to step outside, observe plants, and build environmental awareness.

Taking It Outside

The true test of a nature application is how it performs beyond the desk.

My next step is taking My Garden World out on trails and city parks to test identification across varying daylight conditions, diverse wild flora, and different growth stages.

The goal isn't just an interface that looks nice on a screen, but an experience that sparks genuine curiosity about the living world outside.

Challenges I Faced

• Running Neural Models in the Browser: Compiling and running an 86.6M parameter Vision Transformer on-device required balancing model size and latency. Using INT8 quantization through Transformers.js allowed the model to initialize and classify within milliseconds.
• Decoupling Container Artifacts: Real-world plant photos frequently feature pots, planters, and vases. We built a botanical taxonomy filter to recognize and decouple container objects rather than hallucinating false species classifications.
• State Synchronization: Harmonizing instant client-side Dexie IndexedDB caching with cloud PostgreSQL persistence on Neon to keep user gardens, coins, and plant journals synchronized.

What's Next?

• Expanding the botanical taxonomy database with more regional native flora.
• Introducing seasonal planting calendars and microclimate weather tracking.
• Adding community-shared local nature walks and biodiversity quests.
• Further optimizing WebGPU inference for even faster mobile scans.

My Agent Session

I paired with an AI coding assistant to architect the full-stack system, evaluate open-weight vision transformers, and debug TypeScript and deployment configurations.

Prize Categories

• Open-Source AI
• Touch Grass