Tax Geek - AI Chatbot

This project is a LangChain-powered chatbot app designed to allow users to ask questions related to tax law. It leverages OpenAI embeddings to process and retrieve contextually relevant answers from a collection of tax law documents using a Retrieval-Augmented Generation (RAG) system. Additionally, LangSmith is used to monitor and debug every user interaction, and the system includes pytest tests for ensuring reliable performance.

Features

• RAG System: Combines retrieval from document embeddings and OpenAI's LLM to provide accurate and context-specific answers.
• Vector Embeddings: Automatically processes and stores PDF files into embeddings using FAISS and OpenAI.
• Interactive UI: User-friendly interface built with Streamlit for seamless interaction.
• LangSmith Monitoring: Tracks and analyzes all interactions to improve the chatbot's performance and reliability.
• Test Coverage: Pytest tests ensure the RAG system functions as expected and retrieves accurate results.

Prerequisites

Before running the application, ensure you have the following:

1. Python 3.8 or later
2. Required Python libraries:
   • streamlit
   • faiss-cpu
   • pypdf
   • langchain
   • langchain-openai
   • pytest
   • langsmith

Folder Structure

• ./documentations: Place your tax-related PDF documents here.
• app.py: Main application file.
• create_vectordb.py: Creates vectordb with document embeddings and save the vectordb locally.
• test_rag.py: Contains pytest tests for the RAG system.

How to Run

1. Clone the repository:

   git clone https://github.com/BertrandConxy/Tax-Geek-AI-chatbot.git
   cd Tax-Geek-AI-chatbot

2. Create virtual env

   python -m venv venv

3. Install the dependencies using:

   pip install -r requirements.txt

4. Ensure tax-related PDF documents are in the documentations folder.

5. Create .env file for credentials

   OPENAI_API_KEY
   LANGSMITH_TRACING=True
   LANGSMITH_ENDPOINT
   LANGSMITH_API_KEY
   LANGSMITH_PROJECT
   

6. Run create_vectordb.py to create vector embeddings for the documents and store the db locally.

   python create_vectordb.py
   

7. Run the Streamlit app:

   streamlit run app.py

Demo

Monitoring with LangSmith

LangSmith is integrated into this project to monitor and analyze chatbot interactions. This ensures the app remains robust and user-friendly. To configure LangSmith:

1. Set up your LangSmith account and API key.
2. Ensure the LANGSMITH_API_KEY is added to your environment variables.

Testing the RAG System

Pytest tests are included to validate the functionality of the RAG system. To run the tests:

pytest tests_rag.py