Who wouldn’t want instant answers from their documents? That’s exactly what RAG chatbots do—combining retrieval with AI generation for quick, accurate responses!
In this guide, I’ll show you how to create a chatbot using Retrieval-Augmented Generation (RAG) with LangChain and Streamlit. This chatbot will pull relevant information from a knowledge base and use a language model to generate responses.
I’ll walk you through each step, providing multiple options for response generation, whether you use OpenAI, Gemini, or Fireworks—ensuring a flexible and cost-effective solution.
What is Retrieval-Augmented Generation (RAG)?
RAG is a method that combines retrieval and generation to deliver more accurate and context-aware chatbot responses. The retrieval process pulls relevant documents from a knowledge base, while the generation process uses a language model to create a coherent response based on the retrieved content. This ensures your chatbot can answer questions using the most recent data, even if the language model itself hasn’t been specifically trained on that information.
Imagine you have a personal assistant who doesn’t always know the answer to your questions. So, when you ask a question, they look through books and find relevant information (retrieval), then they summarize that information and tell it back to you in their own words (generation). This is essentially how RAG works, combining the best of both worlds.
In a Flowchart RAG process will somewhat look like this:
Now, let’s get started, and get our own chatbot!
Setting Up the Project Environment
We'll be using Python mostly in this TUTO, if you are JS head you can follow the explanations and go through the documentation of langchain js.
First, we need to set up our project environment. This includes creating a project directory, installing dependencies, and setting up API keys for different language models.
1. Create a Project Folder and Virtual Environment
Start by creating a project folder and a virtual environment:
mkdir rag-chatbot
cd rag-chatbot
python -m venv venv
source venv/bin/activate
2. Install Dependencies
Next, create a requirements.txt file to list all necessary dependencies:
langchain==0.0.329
streamlit==1.27.2
faiss-cpu==1.7.4
python-dotenv==1.0.0
tiktoken==0.5.1
openai==0.27.10
gemini==0.3.1
fireworks==0.4.0
sentence_transformers==2.2.2
Now, install these dependencies:
pip install -r requirements.txt
3. Setting Up API Keys
We’ll be using OpenAI, Gemini, or Fireworks for the chatbot’s response generation. You can choose any of these based on your preferences.
Don't worry if you are experimenting, Fireworks provide $1 worth of API keys for free, and
gemini-1.5-flashmodel is also free to an extent!
Set up a .env file to store the API keys for your preferred model:
# Uncomment your API key
# OPENAI_API_KEY=your_openai_api_key_here
# GEMINI_API_KEY=your_gemini_api_key_here
# FIREWORKS_API_KEY=your_fireworks_api_key_here
Make sure to sign up for these services and get your API keys. Both Gemini and Fireworks offer free tiers, while OpenAI charges based on usage.
Document Processing and Chunking
To give the chatbot context, we’ll need to process documents and split them into manageable chunks. This is important because large texts need to be broken down for embedding and indexing.
1. Create document_processor.py
Create a new Python script called document_processor.py to handle document processing:
from langchain.document_loaders import TextLoader
from langchain.text_splitter import CharacterTextSplitter
class DocumentProcessor:
def __init__(self, file_path):
self.file_path = file_path
def load_and_split(self):
loader = TextLoader(self.file_path)
documents = loader.load()
splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
texts = splitter.split_documents(documents)
return texts
if __name__ == "__main__":
processor = DocumentProcessor("data/sample_text.txt")
texts = processor.load_and_split()
print(f"Processed {len(texts)} text chunks")
This script loads a text file and splits it into smaller chunks of about 1000 characters with a small overlap to ensure that no context is lost between chunks. Once processed, the documents are ready to be embedded and indexed.
Creating Embeddings and Indexing
Now that we have our documents chunked, the next step is to convert them into embeddings (numerical representations of text) and index them for fast retrieval. (as machines understand numbers easier than words)
1. Create embedding_indexer.py
Create another script called embedding_indexer.py:
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
class EmbeddingIndexer:
def __init__(self):
self.embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
def create_vectorstore(self, texts):
vectorstore = FAISS.from_documents(texts, self.embeddings)
return vectorstore
if __name__ == "__main__":
from document_processor import DocumentProcessor
processor = DocumentProcessor("data/sample_text.txt")
texts = processor.load_and_split()
indexer = EmbeddingIndexer()
vectorstore = indexer.create_vectorstore(texts)
print("Vector store created successfully")
In this script, the embeddings are created using a Hugging Face model (all-MiniLM-L6-v2). We then store these embeddings in a FAISS vectorstore, which allows us to quickly retrieve similar text chunks based on a query.
Implementing Retrieval and Response Generation
Here comes the exciting part: combining retrieval with language generation! You’ll now create a RAG chain that fetches relevant chunks from the vectorstore and generates a response using a language model. (vectorstore is a database where we stored our data converted to numbers as vectors)
1. Create rag_chain.py
Let’s create the file rag_chain.py:
from langchain.chains import RetrievalQA
from langchain.llms import OpenAI, Gemini, Fireworks
from dotenv import load_dotenv
import os
load_dotenv()
class RAGChain:
def __init__(self, vectorstore):
self.vectorstore = vectorstore
self.llm = self.get_llm()
def get_llm(self):
if os.getenv("OPENAI_API_KEY"):
return OpenAI(api_key=os.getenv("OPENAI_API_KEY"), temperature=0)
elif os.getenv("GEMINI_API_KEY"):
return Gemini(api_key=os.getenv("GEMINI_API_KEY"), temperature=0)
elif os.getenv("FIREWORKS_API_KEY"):
return Fireworks(api_key=os.getenv("FIREWORKS_API_KEY"), temperature=0)
else:
raise ValueError("No valid API key found! Please set one in .env file.")
def create_chain(self):
retriever = self.vectorstore.as_retriever(search_kwargs={"k": 3})
qa_chain = RetrievalQA.from_chain_type(
llm=self.llm,
chain_type="stuff",
retriever=retriever,
return_source_documents=True
)
return qa_chain
if __name__ == "__main__":
from document_processor import DocumentProcessor
from embedding_indexer import EmbeddingIndexer
processor = DocumentProcessor("data/sample_text.txt")
texts = processor.load_and_split()
indexer = EmbeddingIndexer()
vectorstore = indexer.create_vectorstore(texts)
rag_chain = RAGChain(vectorstore)
qa_chain = rag_chain.create_chain()
query = "What is the capital of France?"
result = qa_chain({"query": query})
print(f"Answer: {result['result']}")
Here, we give you the choice between OpenAI, Gemini, or Fireworks based on the API key you provide. The RAG chain will retrieve the top 3 most relevant documents and use the language model to generate a response.
You can switch between models depending on your budget or usage preferences—Gemini and Fireworks are free, while OpenAI charges based on usage.
Building the Chatbot Interface
Now, we’ll build a simple chatbot interface to take user input and generate responses using our RAG chain.
1. Create chatbot.py
Create a new file called chatbot.py:
class Chatbot:
def __init__(self, qa_chain):
self.qa_chain = qa_chain
def get_response(self, user_input):
try:
response = self.qa_chain({"query": user_input})
return response['result']
except Exception as e:
return f"An error occurred: {str(e)}"
if __name__ == "__main__":
from rag_chain import RAGChain
from document_processor import DocumentProcessor
from embedding_indexer import EmbeddingIndexer
processor = DocumentProcessor("data/sample_text.txt")
texts = processor.load_and_split()
indexer = EmbeddingIndexer()
vectorstore = indexer.create_vectorstore(texts)
rag_chain = RAGChain(vectorstore)
qa_chain = rag_chain.create_chain()
chatbot = Chatbot(qa_chain)
while True:
user_input = input("You: ")
if user_input.lower() in ['exit', 'quit', 'bye']:
print("Chatbot: Goodbye!")
break
response = chatbot.get_response(user_input)
print(f"Chatbot: {response}")
This script creates a command-line chatbot interface that continuously listens for user input, processes it through the RAG chain, and returns the generated response.
Creating the Streamlit UI
It’s time to make your chatbot even more user-friendly by building a web interface using Streamlit. This will allow users to interact with your chatbot through a browser.
1. Create app.py
Create app.py:
import streamlit as st
from document_processor import DocumentProcessor
from embedding_indexer import EmbeddingIndexer
from rag_chain import RAGChain
from chatbot import Chatbot
@st.cache_resource
def initialize_chatbot(file_path):
processor = DocumentProcessor(file_path)
texts = processor.load_and_split()
indexer = EmbeddingIndexer()
vectorstore = indexer.create_vectorstore(texts)
rag_chain = RAGChain(vectorstore)
return Chatbot(rag_chain.create_chain())
st.title("RAG Chatbot")
uploaded_file = st.file_uploader("Upload a text file for the knowledge base", type="txt")
if uploaded_file:
with open("temp_knowledge_base.txt", "wb") as f:
f.write(uploaded_file.getbuffer())
chatbot = initialize_chatbot("temp_knowledge_base.txt")
if "messages" not in st.session_state:
st.session_state.messages = []
for message in st.session_state.messages:
st.chat_message(message["role"]).markdown(message["content"])
if prompt := st.chat_input("Ask a question"):
st.chat_message("user").markdown(prompt)
st.session_state.messages.append({"role": "user", "content": prompt})
response = chatbot.get_response(prompt)
st.chat_message("assistant").markdown(response)
st.session_state.messages.append({"role": "assistant", "content": response})
else:
st.write("Please upload a text file to start chatting!")
2. Run the Streamlit App
To run your Streamlit app, simply use:
streamlit run app.py
This will launch a web interface where you can upload a text file, ask questions, and receive answers from the chatbot.
Optimizing Performance
For better performance, you can experiment with chunk size and overlap when splitting the text. Larger chunks provide more context, but smaller chunks may make retrieval faster. You can also use Streamlit caching to avoid repeating expensive operations like generating embeddings.
If you want to optimize costs, you can switch between OpenAI, Gemini, or Fireworks depending on the query complexity—use OpenAI for complex questions and Gemini or Fireworks for simpler ones to reduce costs.
Wrapping Up
Congratulations! You've successfully created your own RAG-based chatbot. Now, the possibilities are endless:
- Create your own personalized study buddy.
- No more going through long documentations—just "RAG it out" for quick, accurate answers!
The journey starts here, and the potential is limitless!
You can follow my work on GitHub. Feel free to reach out—my DMs are always open on X and LinkedIn.
Top comments (9)
Thank you for this great tutorial! I wrote a blog about how to build a simple serverless chatbot with LangChain last week. Your tutorial is interesting because it adds RAG (not obvious from LangChain's docs but you made it clear and simple), so the chatbot is smarter :)
Thank you for your kind words Qian!
Hello! I’m looking for guidance on downloading and implementing the Create Your Own AI RAG Chatbot using LangChain within Bloxstrap. Any helpful resources or tips would be appreciated. Visit Here
Prettyy Informative!
🙌🙌
Fantastic tutorial! Made building a RAG chatbot with LangChain and Streamlit so clear and accessible.
Thank youu! Hope it helped 🙌
Great article! You made complex concepts feel so beginner-friendly and easy to grasp.
Thank you Mounica!!