📌 INFO-5940

RAG Chatbot

Welcome to the INFO-5940 RAG Chatbot repository! This chatbot is a Retrieval-Augmented Generation (RAG) system powered by OpenAI models and LangChain. It enables users to upload documents (PDF and TXT) and retrieve relevant information from them using semantic search and conversational AI. The chatbot is implemented using Streamlit for the user interface and ChromaDB for vector storage.

---

🛠️ Prerequisites

Before starting, ensure you have the following installed on your system:

• Docker (Ensure Docker Desktop is running)
• VS Code
• VS Code Remote - Containers Extension
• Git
• OpenAI API Key

---

🚀 Setup Guide

1️⃣ Clone the Repository

Open a terminal and run:

git clone https://github.com/cw2236/5940.git
cd 5940

2️⃣ Configure OpenAI API Key

Since docker-compose.yml expects environment variables, follow these steps:

➤ Option 1: Set the API Key in .env (Recommended)

Inside the project folder, create a .env file:

touch .env

Add your API key and base URL:

OPENAI_API_KEY=your-api-key-here
OPENAI_BASE_URL=https://api.ai.it.cornell.edu/
TZ=America/New_York

Make sure the docker-compose.yml include this .env file:

version: '3.8'
services:
  devcontainer:
    container_name: info-5940-devcontainer
    build:
      dockerfile: Dockerfile
      target: devcontainer
    environment:
      - OPENAI_API_KEY=${OPENAI_API_KEY}
      - OPENAI_BASE_URL=${OPENAI_BASE_URL}
      - TZ=${TZ}
    volumes:
      - '$HOME/.aws:/root/.aws'
      - '.:/workspace'
    env_file:
      - .env

Compose the container:

docker-compose up --build

Now, your API key will be automatically loaded inside the container.

---

3️⃣ Open in VS Code with Docker

1. Open Docker dashboard and run the imange you just created(It should be called 5940)

2. Open VS Code and navigate to the project folder.

3. Open the Command Palette (Ctrl+Shift+P or Cmd+Shift+P on Mac) and search for:

   Remote-Containers: Rebuild and Reopen in Container"
   

4. Select From 'docker-compose.yml', then don't select any choice and click OK directly

5. Select this option. VS Code will build and open the project inside the container.

📌 Note: If you don’t see this option, ensure that the Remote - Containers extension is installed.

---

4️⃣ Install Additional Libraries

pip install -U langchain-chroma
pip install pdfminer.six

---

5️⃣ Running the Chatbot

Once the setup is complete, run the chatbot with the following command:

streamlit run Chatbot_A1.py

---

Tests Conducted

Results of the tests follow the logic and step defined in the prompt template.

• Tested with asking for domain specific questions such as "Who is Ayham Boucher?" (WITHOUT uploading any documents)
• Tested with asking for general questions such as "What is a banana?" , "Who is the first president of the U.S.?"(WITHOUT uploading any documents)
• Tested with asking for general questions such as "What is a banana?", "Who is the first president of the U.S.?" (AFTER uploading single/multiple document(s))
• Tested with domain specific questions such as "Who is Ayham Boucher", "What does Ayham teach?", "What is Harvard's school color?", "When was Cornell built?" (AFTER uploading single/multiple document(s))

If you hope to start over after testing, please feel free to delete the 'uploaded_docs' and 'chroma_db' directories and re-run the AI chatbot.

📌 Features

• Upload and process PDF and TXT documents.
• Store document embeddings using ChromaDB.
• Retrieve relevant document sections using similarity search.
• Generate responses using OpenAI's GPT-4o model.
• Maintain conversation history within a session.
• Able to identify source of information.
• Answer the question even when context does not include useful information, with notification to users that response is generated not based on the context, or respond with "I don't know".

---

📂 File Upload Processing

• Uploaded files are saved in the uploaded_docs directory.
• Documents are split into chunks for efficient retrieval.
• Chunks are stored in ChromaDB for similarity search, along with metadata (source file name).

---

📌 RAG Pipeline

1. Document Upload: Users upload PDF/TXT files.
2. Text Processing: Extracted text is chunked using RecursiveCharacterTextSplitter.
3. Vector Storage: Chunks and source of information are embedded and stored in ChromaDB.
4. Retrieval: Similar document chunks and corresponding metadata about source files are retrieved based on user queries.
5. Generation: Retrieved content is used to generate an answer using GPT-4o, following tailored prompts.

---

🙏 Acknowledgement

This project is developed as part of INFO-5940 at Cornell University.