Build Your First AI Chatbot: RAG with NVIDIA

Organized by: ACM-W

This repository contains the complete code used in the ACM-W workshop “Build Your First AI Chatbot: RAG with NVIDIA”.
In this workshop, students learn the basics of Retrieval-Augmented Generation (RAG) and build a working AI chatbot that can answer questions from their own documents.

---

Basics of RAG (Retrieval-Augmented Generation)

RAG is a technique that combines document retrieval with text generation.

Instead of answering only from what a model already knows, a RAG system:

• First searches relevant information from uploaded documents
• Then uses that information as context to generate accurate answers

This makes the chatbot:

• More accurate
• Less likely to hallucinate
• Able to answer from private PDFs and documents
• Easy to update without retraining the model

---

What This Chatbot Does

• Allows users to upload TXT or PDF documents
• Splits documents into smaller chunks
• Converts each chunk into embeddings using NVIDIA AI
• Stores embeddings in a vector database using ChromaDB
• Retrieves the most relevant chunks for each question
• Sends the retrieved context to the LLM
• Generates grounded answers from the documents
• Maintains chat history during the session

---

Tech Stack

• Python
• Streamlit
• ChromaDB
• NVIDIA NIM
• Requests
• PyPDF2
• python-dotenv

---

How to Run the Project

Clone the github repository

git clone <PASTE YOUR GITHUB REPO LINK HERE>

Go inside the project folder

cd <YOUR REPO FOLDER NAME>

Install all required python packages

pip install streamlit requests chromadb PyPDF2 python-dotenv

Create a file to store your nvidia api key (.env) & add your nvidia api key to the .env file

NVIDIA_API_KEY=nvapi-xxxxxxxxxxxxxxxx

Run the streamlit application

streamlit run main.py

Important Notes:

• The vector database is created locally on your system
• You can upload multiple files and the chatbot will answer using all of them
• To reset everything, delete the chroma folder or use the clear option if available