HR Policy Chatbot

🚀 RAG Chatbot with Local MCP Server

Build an RAG-based HR chatbot using a localhost MCP server as the function call center to provide workplace rules

🚀 Quick Start

This project implements a Retrieval Augmented Generation (RAG) chatbot based on Streamlit and an MCP server. Users can upload PDF files, and the chatbot extracts relevant information from these PDFs to answer natural language questions. The system utilizes OpenAI models, LangChain tools, and an in-memory vector store for efficient document retrieval.

✨ Features

• MCP Tool Integration: Use MCP for tool orchestration to ensure smooth communication between document indexing, retrieval, and answer generation. Backend tools can be easily extended or replaced to add new features.

• PDF Upload and Parsing: After uploading PDF files, use PDFPlumberLoader to extract text content.

• Text Chunking: Use RecursiveCharacterTextSplitter to split the extracted text into smaller paragraphs for easier indexing and retrieval.

• Document Indexing: Index the chunked text into an in-memory vector store and generate embeddings using OpenAIEmbeddings.

• Similarity Search (Cosine Similarity): When a user submits a query, the chatbot performs a similarity search to retrieve the document paragraphs most relevant to the query.

• Prompt-based Answer Generation: Use a custom prompt template to combine the user's question and the retrieved context, and generate the final answer through a GPT-4-powered LLM (using ChatOpenAI).

• Interactive Interface: The application uses Streamlit to provide an interactive chat interface that displays user questions and robot answers.

📚 Documentation

Overview

This project realizes a Retrieval Augmented Generation (RAG) chat - based on Streamlit and an MCP server. Users can upload PDF files, and the chatbot extracts relevant information from these PDFs to answer natural language questions. The system leverages OpenAI models, LangChain tools, and an in - memory vector store for efficient document retrieval.

Functionality Details

• MCP Tool Integration: MCP is used for tool orchestration, ensuring seamless communication among document indexing, retrieval, and answer generation. Back - end tools can be easily extended or replaced to incorporate new features.
• PDF Upload and Parsing: Once PDF files are uploaded, PDFPlumberLoader is employed to extract text content.
• Text Chunking: The RecursiveCharacterTextSplitter is used to divide the extracted text into smaller paragraphs for easier indexing and retrieval.
• Document Indexing: The chunked text is indexed into an in - memory vector store, and OpenAIEmbeddings are used to generate embeddings.
• Similarity Search (Cosine Similarity): When a user submits a query, the chatbot conducts a similarity search to retrieve the most relevant document paragraphs.
• Prompt - based Answer Generation: A custom prompt template is used to combine the user's question and the retrieved context, and the final answer is generated via a GPT - 4 - powered LLM (using ChatOpenAI).
• Interactive Interface: The application offers an interactive chat interface using Streamlit, which displays user questions and robot answers.

Results

[Since the original "Results" section has no content, this section is skipped according to the rules.]