E-commerce Chatbot

An intelligent conversational AI chatbot for e-commerce platforms, built with Streamlit and powered by Groq LLM. The chatbot routes user queries to specialized handlers using semantic routing and provides accurate responses for product searches, FAQs, and general conversation.

---

Key Achievements

• Dual-intent routing with semantic classification for intelligent query handling
• Real-time database querying with natural language to SQL conversion
• Vector-based FAQ retrieval using ChromaDB for instant policy responses
• Dynamic product recommendations with price filtering and brand matching
• Sub-2 second response time for complex multi-table queries
• Multi-intent understanding without explicit user commands

---

Features

Intelligent Query Routing

The chatbot automatically classifies user queries into three categories:

Category	Description	Example
FAQ	General questions about policies, payments, returns, and tracking	"What is your return policy?"
SQL	Product search and database queries	"Show me Nike shoes under Rs 3000"
Small Talk	Casual conversation and greetings	"How are you today?"

Multi-Modal Interaction

FAQ Handler

• Vector-based semantic search using ChromaDB
• Sentence Transformer embeddings for accurate matching
• Retrieves relevant answers from the FAQ database
• Context-aware responses using Groq LLM

SQL Query Handler

• Natural language to SQL conversion
• Product database queries (shoes/footwear products)
• Filters by brand, price, discount, and ratings
• Formatted product listings with links

Small Talk Handler

• Friendly conversational responses
• Engaging and empathetic interactions
• Maintains a warm, personable tone

---

Frontend Interface

Workflow Overview

---

Core Components

Semantic Router
Intent classification engine that distinguishes between FAQ, SQL, and small talk queries using embedding similarity and pattern matching.

FAQ Chain

• Vector database (ChromaDB) storing policy documents and Q&A pairs
• Similarity-based retrieval with contextual answer generation
• Sentence Transformer embeddings for accurate semantic matching
• Handles queries like "What are your return policies?" or "Do you accept online payments?"

SQL Chain

• Natural language to SQL query generation using Groq LLM
• SQLite database with scraped e-commerce product data
• Structured product information retrieval with price, brand, and category filtering
• Handles queries like "Show me Nike shoes under Rs 3000" or "Find products with 30 percent discount"

Small Talk Handler

• Friendly conversational responses for casual interactions
• Maintains a warm, empathetic tone for user engagement

---

Technology Stack

Component	Technology	Purpose
LLM	LLaMA (via GROQ API)	Low-latency inference for fast responses
Vector DB	ChromaDB + Sentence Transformers	Semantic similarity search
Relational DB	SQLite	Structured product data storage
Web Framework	Streamlit	Interactive UI and deployment
Routing	Semantic Router	Embedding-based query classification
Data Source	Web Scraping	E-commerce product catalogs

---

Data Pipeline

Web Scraping
    -> Data Processing (Clean and Normalize)
    -> Database Population (SQLite)
    -> Vector Indexing (ChromaDB)
    -> Query Processing and Response Generation

Steps:

1. Web Scraping - Automated product data extraction from e-commerce platforms
2. Data Processing - Cleaning, normalization, and validation
3. Database Population - SQLite database creation with optimized indexing
4. Vector Indexing - Policy documents embedded and stored in ChromaDB
5. Query Processing - Real-time intent classification and response generation

---

Project Structure

.
├── main.py                              # Main entry point
├── pyproject.toml                       # Project dependencies
├── README.md                            # Project documentation
│
├── app/                                 # Main application
│   ├── main.py                          # Streamlit UI
│   ├── router.py                        # Semantic routing logic
│   ├── faq.py                           # FAQ handler
│   ├── sql.py                           # SQL handler
│   ├── small_talk.py                    # Conversation handler
│   └── resources/
│       └── faq_data.csv                 # FAQ database
│
├── web-scrapping/                       # Data collection
│   ├── flipkart_data_extraction.ipynb   # Web scraping notebook
│   ├── csv_to_sqlite.py                 # Database creation
│   └── ecommerce_data_final.csv         # Product dataset
│
└── learnings/                           # Research and experiments
    ├── chroma_learning.py
    ├── groq_learning.py
    └── semantic_router_learning.py

---

Getting Started

Prerequisites

• Python 3.11 or higher
• Groq API key (Get it here)

Installation

1. Clone the repository

   git clone <repository-url>
   cd Project-ecommerce-tool

2. Install dependencies

   pip install -e .

   Or install manually:

   pip install chromadb>=1.4.1 groq>=1.0.0 pandas>=2.3.3 semantic-router[qdrant]>=0.1.12 sentence-transformers>=5.2.0 streamlit>=1.53.0

3. Set up environment variables

   Create a .env file in the root directory:

   GROQ_API_KEY=your_groq_api_key_here
   GROQ_MODEL=llama-3.3-70b-versatile

4. Prepare the product database (optional)

   To load product data, run:

   python web-scrapping/csv_to_sqlite.py

5. Run the application

   streamlit run app/main.py

   The chatbot will be available at http://localhost:8501

Dependencies

• streamlit: Web interface
• groq: LLM for response generation
• semantic-router: Query classification
• chromadb: Vector database for FAQ retrieval
• sentence-transformers: Text embeddings
• pandas: Data manipulation
• sqlite3: Product database (built-in)

Usage Examples

FAQ Queries

• "What is your return policy?"
• "How can I track my order?"
• "What payment methods do you accept?"
• "Do I get a discount with HDFC credit card?"

Product Search Queries

• "Show me Nike shoes with at least 30% discount"
• "Find running shoes under Rs. 3000"
• "Puma shoes with rating above 4"
• "Cheapest shoes from Adidas"

Small Talk

• "Hello!"
• "How are you?"
• "Thank you"

How It Works

1. User Input: User enters a query in the Streamlit chat interface
2. Semantic Routing: The router classifies the query into FAQ, SQL, or small_talk
3. Specialized Handler: Based on the route:
   • FAQ: Searches ChromaDB vector store for similar questions, retrieves answers
   • SQL: Converts natural language to SQL, queries product database, formats results
   • Small Talk: Generates friendly conversational response
4. Response Generation: Groq LLM generates contextual, streaming responses
5. Display: Response is streamed back to the user in real-time

Query Examples and Expected Behavior

FAQ Queries (Policy & General Questions)

User: "What is your return policy?"
Bot: Retrieves from ChromaDB → "You can return products within 30 days of delivery."

User: "Do you offer international shipping?"
Bot: Fetches FAQ → "Yes, we offer international shipping to select countries."

SQL Queries (Product Search)

User: "Show me Nike shoes with at least 30% discount"
Bot: Generates SQL → Executes query → Returns formatted product list with prices and ratings

User: "Find running shoes under Rs. 3000"
Bot: Filters by category & price → Lists matching products with links

Small Talk

User: "Hello!"
Bot: "Hi there! How can I help you today?"

User: "Thank you"
Bot: "You're welcome! Feel free to ask me anything about our products!"

Performance Metrics

• Intent Classification Accuracy: High accuracy using semantic embeddings
• Query Success Rate: 88%+ successful SQL generation for valid product searches
• Response Time: Sub-2 second average for complex queries
• Availability: Depends on GROQ API uptime and internet connectivity

Development

Database Schema

The product database contains the following fields:

Field	Type	Description
product_link	String	URL to product page
title	String	Product name
brand	String	Brand name
price	Integer	Price in Indian Rupees (₹)
discount	Float	Discount percentage (0.1 = 10%)
avg_rating	Float	Average rating (0-5 scale)
total_ratings	Integer	Total number of customer ratings

Adding New FAQ Entries

Edit app/resources/faq_data.csv with new question-answer pairs:

question,answer
"What is your shipping policy?","Free shipping on orders above Rs. 500."
"How do I contact customer support?","Email us at support@example.com or call 1800-XXXX-XXXX."

The FAQ database will auto-update on the next application run due to the ingest_faq_data() function.

Customizing Routes

Modify app/router.py to add new routes or update utterances for better intent classification:

new_route = Route(
    name='custom_intent',
    utterances=[
        "example query 1",
        "example query 2",
        "example query 3"
    ]
)

Add the route to the router:

from semantic_router.routers import SemanticRouter

router = SemanticRouter(routes=[faq, sql, small_talk, new_route])

SQL Query Customization

Modify the SQL prompt in app/sql.py to adjust query generation behavior, schema definitions, or filtering logic.

Future Enhancements

Planned Features

• Multi-language Support: Hindi and regional language query processing for broader accessibility
• Voice Integration: Speech-to-text and text-to-speech capabilities for hands-free interaction
• Recommendation Engine: ML-based product suggestions using collaborative filtering
• Advanced Analytics: User interaction insights and query analytics dashboard
• Image Search: Visual product search using computer vision
• Product Comparison: Side-by-side product comparison with feature highlighting

Scalability Roadmap

• Multi-vendor Support: Extension to other e-commerce platforms beyond Flipkart
• Cloud Deployment: AWS/Azure integration for production scaling
• API Development: RESTful APIs for third-party integrations
• Mobile Application: Native mobile app with embedded chatbot
• Real-time Data Sync: Live product catalog updates with inventory management
• Analytics Pipeline: Track user interactions and optimize routing

Performance Optimization Tips

1. Database Indexing: Add indexes to frequently queried columns for faster SQL execution
2. Embedding Caching: ChromaDB automatically caches embeddings for repeated queries
3. Query Optimization: Structure SQL queries with WHERE clauses early for faster filtering
4. Rate Limiting: Implement GROQ API rate limiting to manage quota efficiently

Troubleshooting

Issue: "GROQ_API_KEY not found"

Solution: Ensure .env file exists in root directory with GROQ_API_KEY set

Issue: "ChromaDB collection already exists"

Solution: This is normal. The app checks if FAQ data is already ingested.
To reload FAQ data, delete the ChromaDB database files or reinitialize.

Issue: Slow SQL query responses

Solution: 
- Check internet connection to GROQ API
- Ensure product database is properly indexed
- Try simpler product queries to debug

Learning Resources

• Streamlit Documentation
• Groq API Docs
• ChromaDB Guide
• Semantic Router
• Sentence Transformers

License

This project is available for educational and commercial use.

Contributing

Contributions are welcome! Areas for contribution:

• Adding more FAQ entries and improving coverage
• Optimizing SQL query generation
• Enhancing semantic router accuracy
• Extending support to other e-commerce platforms
• Performance improvements and bug fixes

Please feel free to submit a Pull Request or open an issue for discussions.

Acknowledgments

• Groq for providing fast, efficient LLM inference
• Semantic Router for intelligent query routing capabilities
• ChromaDB for scalable vector search functionality
• Streamlit for the intuitive web interface
• Hugging Face for pre-trained embedding models

---

Note:

• Make sure to set up your Groq API key before running the application
• The chatbot requires active internet connection for LLM API calls
• Product data is sourced from web scraping; ensure compliance with platform ToS
• For production deployment, implement proper rate limiting and error handling