Florabyte – Project Documentation

Florabyte is a Next.js app that helps users analyze plant photos using AI, generate care plans, and discover relevant gardening products. It also supports shopping cart checkout via Midtrans and a lightweight admin dashboard for order management.

Tech Stack

• Framework: Next.js 15 (App Router, Turbopack)
• Language: TypeScript
• UI: Tailwind CSS, Radix UI, lucide-react
• Auth: NextAuth (Google OAuth)
• DB: MongoDB (native driver), Atlas Vector Search for product recommendations
• Media: Cloudinary (image storage)
• AI: Google Generative AI (Gemini Vision + text-embedding-004)
• Notifications: react-hot-toast

Key Features

• Plant analysis: upload a plant photo, AI returns identification, care plan, and recommended products
• Product discovery: vector search over product embeddings using AI-generated keywords
• Shopping cart + Midtrans payments (status reflected on cart document)
• Admin: view and update order status lifecycle
• User profile: Google OAuth sign-in and account actions

Repository Structure (high-level)

• src/app: App Router pages and API routes
  • plants/scan: image upload + result UI
  • api/plants/analyze: AI + vector recommendations
  • api/payment/create|notification: Midtrans integration
  • admin/orders: admin page + API
  • products, cart, profile, login, register
• src/db: Mongo config and domain models (Cart, Product, User, etc.)
• src/types: Shared TypeScript types (ProductDoc, ProductRecommendation, Plant types)
• src/lib: Auth config, utilities
• public/: Static assets

Environment Variables

Create a .env file with the following variables (example names):

• App/Next.js

  • NEXTAUTH_URL – external URL for NextAuth
  • NEXTAUTH_SECRET – NextAuth secret

• Google OAuth

  • GOOGLE_CLIENT_ID
  • GOOGLE_CLIENT_SECRET

• Google AI

  • GOOGLE_API_KEY – used for Gemini Vision and Embeddings
  • GEMINI_MODEL – e.g. gemini-1.5-pro

• MongoDB

  • MONGODB_URI – connection string
  • DB_NAME – database name

• Cloudinary

  • CLOUDINARY_CLOUD_NAME
  • CLOUDINARY_API_KEY
  • CLOUDINARY_API_SECRET

• Midtrans

  • MIDTRANS_SERVER_KEY
  • MIDTRANS_CLIENT_KEY
  • MIDTRANS_BASE_URL – sandbox/prod endpoint

Getting Started

1. Install dependencies
   • npm install
2. Run dev server
   • npm run dev
3. Open the app
   • http://localhost:3000

Build & Run

• Production build: npm run build
• Start: npm start
• Lint: npm run lint

Core Flows

1. Plant Analysis & Recommendations

• UI: src/app/plants/scan/page.tsx
• API: src/app/api/plants/analyze/route.ts
  • Accepts an image via multipart form-data
  • Calls Gemini Vision with a constrained JSON prompt
  • Extracts keywords + supplies + species
  • For each term, generates an embedding and performs Mongo $vectorSearch on Products.embedding
  • Aggregates highest score per product, returns top 8
  • No regex/stock fallback (by design) – may return empty array

Product schema must contain an embedding: number[] and a configured Atlas Vector Search index (products_vector_index) pointing to that field.

2. Cart, Checkout & Orders

• Cart stored as a document with fields like productIds, total, status, orderStatus, and midtransOrderId
• Payment creation: POST /api/payment/create
• Midtrans notifications: POST /api/payment/notification
• Admin order management:
  • List: GET /api/admin/orders
  • Status update: PATCH /api/orders/:id/status

3. Authentication

• NextAuth configured in src/lib/auth.ts with GoogleProvider
• After sign-in, users land on /profile

Data Models (selected)

• ProductDoc:
  • _id, name, description, price, stock, imgUrl, category, embedding: number[]
• ProductRecommendation (API response item):
  • _id, name, description, price, stock, imgUrl, category, score
• Cart (simplified order):
  • status (pending | paid | cancelled)
  • orderStatus (Pending | Diproses | Dikirim | Selesai | Dibatalkan)
  • productIds (duplicate ObjectIds represent quantity)
  • total, midtransOrderId, timestamps

Indexing & Search Notes

• Ensure a MongoDB Atlas Vector Search index exists on Products.embedding with name products_vector_index
• $vectorSearch cannot filter on fields that aren’t indexed. This route executes the vector stage first, then filters (e.g., stock > 0) in a $match stage.

Development Tips

• Add product embeddings at import time or via a batch job so $vectorSearch returns meaningful results.
• Keep AI prompt concise; route sanitizes responses to valid JSON only.
• For debugging recommendations, the analyze route returns terms and a diagnostics object (e.g., per-term hit counts).

Security & Secrets

• Never commit .env values.
• Midtrans webhook: validate signatures and add idempotency as needed.
• Ensure Cloudinary keys have the minimum required permissions.

Troubleshooting

• Empty recommendations: verify embedding exists on products and the Atlas index name matches.
• Vector search filter errors: avoid using $vectorSearch.filter on non-indexed fields; this code matches after the vector stage.
• Build/type errors: run npm run build and fix reported TypeScript issues.

License

Proprietary – for internal use within the project/team unless stated otherwise.