Customer Support Chatbot

Overview

This project is a full-stack chatbot solution. The aim is to optimize sales processes for companies in construction and home materials. The chatbot interacts with a product catalog dataset and improves the customer experience by providing relevant, context-driven responses.

Two integration approaches were explored for the chatbot:

1. Embedding Database: Used to fetch relevant context from the CSV but had limitations in performance and flexibility.
2. LangChain: Chosen as the primary framework for connecting the Language Learning Model (LLM) to the product catalog dataset, enabling dynamic context retrieval, chat history integration, and prompt refinement.

Table of Contents

• Installation
• Setup
• Running the Project
• Provided Resources
• Task Implementation
• Environment Variables
• Functionality Overview
• Contributing

Installation

1. Clone the Repository

git clone https://github.com/amirzarandi/internship-case-study
cd internship-case-study

2. Install Python and Pipenv

• Ensure Python is installed on your system.
• Install Pipenv for virtual environment management:

  pip install pipenv

3. Install Dependencies

• Create and activate the virtual environment using Pipenv:

  pipenv install
  pipenv shell

• This will install all the required dependencies for the backend.

Setup

1. Backend Setup

• The backend uses Flask to create an LLM-powered chat endpoint connected to the product catalog dataset.
• To set up the backend, run:

  flask run

• Make sure to set up the environment variable OPENAI_API_KEY for LLM integration (see Environment Variables below).

2. Database Setup

• The provided Python scripts initialize a database storing product catalog data from the CSV file.
• To set up the database, run:

  python add_csv_knowledge_langchain.py
  python add_csv_knowledge_embed.py

3. Frontend Setup

• The frontend is based on a React boilerplate template provided by Instalily.
• You can run it using npm (Node Package Manager):

  npm install
  npm run

• The frontend has been modified to enhance user experience and visuals, providing better interaction with the chatbot.

Running the Project

1. Start the Backend

• Ensure the virtual environment is active, and the OPENAI_API_KEY is set:

  flask run

2. Start the Frontend

• From the frontend directory, start the React application:

  npm run

Provided Resources

This case study includes:

1. Product Catalog Dataset: A CSV file containing product details that the chatbot uses for context.
2. Frontend Boilerplate Template: A React-based template for the chat interface, which has been modified to enhance user experience.

Task Implementation

The task was to build a full-stack chatbot solution with the following criteria:

1. Backend Chat Endpoint:
   • Developed using Flask and LangChain to connect the product catalog CSV data to the LLM.
   • LangChain dynamically retrieves context from the CSV, incorporates chat history, and refines prompts based on user input.
2. Integrated Frontend:
   • The frontend was customized from the provided boilerplate to improve the user interface, interaction flow, and visual design.
   • User-friendly elements were added to make the chat experience more intuitive.

Environment Variables

• OPENAI_API_KEY: Required for interacting with the OpenAI LLM.
  • Set this variable in your terminal or environment file before running the backend:

    export OPENAI_API_KEY='your_openai_api_key'

Functionality Overview

1. LangChain-based Contextual Interaction

• The chatbot retrieves context from the CSV using LangChain, allowing it to process user inputs effectively.
• The process includes:
  1. Context Retrieval: Fetches up to 10 relevant rows related to the user’s query.
  2. History Integration: Adds chat history to maintain continuity over multiple interactions.
  3. User Input: Combines the input with context to refine the final prompt.
  4. LLM Querying: Queries the LLM for a refined response.

2. Embedding Database (Initial Approach)

• Initially, embeddings were used for context retrieval from the CSV, but this approach was slower and less flexible than LangChain.

Contributing

Feel free to open issues or submit pull requests for improvements, bug fixes, or new features.