Chukde RAG Chatbot Setup Guide

Prerequisites

• Docker and Docker Compose
• PHP 8.1+ with Composer
• OpenAI API key

Quick Setup

1. Create environment file:

   cp .env.example .env
   # Or run: bash setup-env.sh

2. Update your OpenAI API key in .env:

   OPENAI_API_KEY=your-actual-openai-api-key
   

3. Start services:

   docker compose up -d

4. Install dependencies and setup Laravel:

   composer install
   php artisan key:generate

5. Run migrations and seed data:

   php artisan migrate
   php artisan db:seed --class=DemoProductSeeder

6. Index products in Qdrant:

   php artisan rag:reindex

7. Start the application:

   php artisan serve

Testing the API

Test with exact SKU match:

curl -X POST http://127.0.0.1:8000/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"q":"KRC-200"}'

Test with semantic search:

curl -X POST http://127.0.0.1:8000/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"q":"Kashmiri Red Chilli 200g price"}'

Expected Response

Success response:

{
  "reply": "Here are the closest matches:",
  "products": [
    {
      "id": 1,
      "sku": "KRC-200",
      "name": "Kashmiri Red Chilli 200g",
      "brand": "Chukde",
      "price": 220.00,
      "offer_price": 199.00,
      "currency": "INR",
      "price_updated_at": "2025-01-21 12:01:33",
      "url": "https://chukde.com/product/kashmiri-red-chilli-200g"
    }
  ]
}

Not found response:

{
  "reply": "I couldn't find that product on chukde.com right now."
}

Architecture

• MySQL: Stores product data and live prices
• Qdrant: Vector database for semantic search
• OpenAI: Embeddings API for text vectorization
• Laravel: API framework with auto-indexing via model events

Auto-Indexing

Products are automatically indexed when:

• Product is created/updated (UpsertProductVector job)
• Product is deleted (DeleteProductVector job)

In development, jobs run synchronously (QUEUE_CONNECTION=sync). In production, use queue workers with QUEUE_CONNECTION=database or redis.

Tuning

• Adjust score_threshold in ChatController (default: 0.75)
• Higher threshold = more strict matching
• Lower threshold = more lenient matching