Author details

Swathi Radhakrishnan
AI Developer
GitHub Repository

Abstract

This project presents a Retrieval-Augmented Generation (RAG) based intelligent assistant designed for answering Frequently Asked Questions (FAQs) on an e-commerce platform. Leveraging LangChain, Groq’s LLaMA model, Chroma vector database, and HuggingFace embeddings, the assistant delivers fast, accurate, and context-aware responses to customer queries. Trained on real-world FAQ data from The Good Guys e-commerce platform, the system improves the customer support experience by enabling natural language interaction with business-specific information.

Introduction

Traditional FAQ pages often require users to manually search through lists of static content. This project aims to enhance the customer support experience by implementing a conversational interface that provides semantic, context-aware responses using Retrieval-Augmented Generation (RAG). By integrating LangChain and Groq’s high-speed LLMs with a vector search backend, users can engage in fluid, natural interactions to resolve queries about orders, deliveries, and store policies.

Methodology

2.1 Data Collection and Processing

• The dataset comprises FAQ pairs (questions and answers) scraped from The Good Guys e-commerce website.
• The data is stored in a CSV format with two columns: question and answer.

2.2 Embedding and Vector Storage

• Text embeddings are generated using HuggingFace sentence-transformers.
• These embeddings are stored in a Chroma vector database for fast semantic retrieval.

2.3 RAG Pipeline

• When a user asks a question:
  • The query is embedded and matched against stored vectors.
  • The top relevant context is retrieved.
  • The Groq LLaMA model uses this context to generate a coherent answer.
• The pipeline is orchestrated using LangChain.

2.4 Frontend Interface

• A Streamlit application provides a simple chat interface.
• Features include:
  • Real-time Q&A
  • Conversational memory
  • Display of prior interactions
  • UI snapshot included (assets/ui_preview.png)

Experiments

3.1 Evaluation Setup

• The assistant was tested using common customer support queries.
• Example domains included:
  • Online orders
  • Delivery services
  • Click and Collect
  • Store Cash usage

3.2 Sample Queries

• “How do I track my online order?”
• “How does Click and Collect work?”
• “What are the delivery charges?”
• “Can I use store cash for online purchases?”

3.3 Deployment Environment

• Local testing was conducted via:

  streamlit run src/app.py
  

• Requires GROQ_API_KEY and data file in /data folder

Results

• Accuracy: The assistant reliably retrieved the correct context and generated answers with high relevance to the user query.

• Latency: Fast response generation was observed using Groq’s LLaMA model, significantly outperforming traditional cloud-hosted LLMs in terms of speed.

• User Experience: The natural conversational interface made the assistant easy to use even for non-technical users.

Conclusion

The E-commerce FAQ Assistant demonstrates a practical use case for combining vector search with LLMs to improve customer experience in online retail. The system performs well across a range of common questions, showcasing the power of Retrieval-Augmented Generation in domain-specific applications.

Future Enhancements

• Multilingual support
• Voice-based interaction
• Broader dataset coverage

This work serves as a strong foundation for deploying conversational AI agents tailored to specific business knowledge bases.