How to Build a QA System with LangChain: Simple Guide

To build a QA system with LangChain, you create a chain that connects a language model with a document retriever to answer questions based on your data. Use RetrievalQA to combine a retriever and a language model for easy question answering.

Syntax

The basic syntax to build a QA system in LangChain involves these parts:

• Language Model: The AI model that generates answers (e.g., OpenAI's GPT).
• Retriever: Finds relevant documents or data to answer the question.
• RetrievalQA Chain: Combines the retriever and language model to produce answers.

Example syntax:

from langchain.chains import RetrievalQA
from langchain.llms import OpenAI

llm = OpenAI()
retriever = your_retriever_here
qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)

python

from langchain.chains import RetrievalQA
from langchain.llms import OpenAI

llm = OpenAI()
retriever = your_retriever_here
qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)

Example

This example shows how to build a simple QA system using LangChain with OpenAI and a vector store retriever. It loads documents, creates embeddings, and answers a question.

python

from langchain.chains import RetrievalQA
from langchain.llms import OpenAI
from langchain.vectorstores import FAISS
from langchain.embeddings.openai import OpenAIEmbeddings

# Sample documents
texts = ["LangChain helps build language model apps.", "You can create QA systems easily."]

# Create embeddings for documents
embeddings = OpenAIEmbeddings()
vectorstore = FAISS.from_texts(texts, embeddings)

# Create retriever from vectorstore
retriever = vectorstore.as_retriever()

# Initialize language model
llm = OpenAI()

# Build QA chain
qa = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)

# Ask a question
query = "What does LangChain help with?"
answer = qa.run(query)
print(answer)

Output

LangChain helps build language model apps.

Common Pitfalls

Common mistakes when building a QA system with LangChain include:

• Not initializing the retriever properly, causing no documents to be found.
• Using a language model without API keys or incorrect configuration.
• Passing raw text instead of embeddings to the vector store.
• Ignoring asynchronous calls if using async APIs.

Example of a wrong retriever setup and the correct way:

python

from langchain.llms import OpenAI
from langchain.chains import RetrievalQA

# Wrong: retriever is just a list of texts (incorrect)
texts = ["Doc1", "Doc2"]
retriever = texts  # This will cause errors

# Correct: use a vector store retriever
from langchain.vectorstores import FAISS
from langchain.embeddings.openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings()
vectorstore = FAISS.from_texts(texts, embeddings)
retriever = vectorstore.as_retriever()

llm = OpenAI()
qa = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)

Quick Reference

Tips for building QA systems with LangChain:

• Always create embeddings for your documents before using a vector store.
• Use RetrievalQA.from_chain_type to combine retriever and LLM easily.
• Make sure your API keys for OpenAI or other LLMs are set in environment variables.
• Test your retriever separately to ensure it returns relevant documents.

Key Takeaways

Use RetrievalQA to connect a retriever and language model for question answering.

Create embeddings and a vector store retriever to find relevant documents.

Ensure your language model is properly configured with API keys.

Test retriever outputs before integrating with the QA chain.

Avoid passing raw texts directly as retrievers; always use vector stores or similar.