Code beginner · 3 min read

How to load PDF for RAG in python

Q: How to load PDF for RAG in python

Use LangChain's PyPDFLoader to load PDF documents, then embed and index them with a vector store like FAISS for RAG workflows in Python.

Direct answer

Use LangChain's PyPDFLoader to load PDF documents, then embed and index them with a vector store like FAISS for RAG workflows in Python.

Setup

Install

bash

pip install langchain_openai langchain_community faiss-cpu PyPDF2

Env vars

OPENAI_API_KEY

Imports

python

from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
import os

Examples

inLoad a single PDF file 'example.pdf' and create a FAISS vector store.

outLoaded 10 pages from example.pdf and created FAISS index with 10 vectors.

inLoad multiple PDFs from a folder and build a combined vector store for RAG.

outLoaded 50 pages from 5 PDFs and indexed 50 vectors for retrieval.

inLoad a large PDF with scanned images (unsupported by PyPDFLoader).

outPyPDFLoader failed to extract text; consider OCR preprocessing before loading.

Integration steps

Install required packages and set your OPENAI_API_KEY in environment variables.
Use PyPDFLoader to load and split the PDF into documents.
Initialize OpenAIEmbeddings to convert documents into vectors.
Create a FAISS vector store from the embedded documents.
Use the vector store for similarity search in your RAG pipeline.

Full code

python

import os
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader

# Load PDF file
pdf_path = "example.pdf"
loader = PyPDFLoader(pdf_path)
docs = loader.load()

print(f"Loaded {len(docs)} pages from {pdf_path}.")

# Initialize embeddings
embeddings = OpenAIEmbeddings(api_key=os.environ["OPENAI_API_KEY"])

# Create FAISS vector store
vector_store = FAISS.from_documents(docs, embeddings)

print(f"Created FAISS index with {len(docs)} vectors.")

# Example similarity search
query = "What is the main topic of the document?"
results = vector_store.similarity_search(query, k=3)
for i, doc in enumerate(results, 1):
    print(f"Result {i}: {doc.page_content[:200]}...")

output

Loaded 10 pages from example.pdf.
Created FAISS index with 10 vectors.
Result 1: The main topic of the document is Retrieval-Augmented Generation (RAG), which combines...
Result 2: RAG workflows typically involve embedding documents and querying them with...
Result 3: This document explains how to use LangChain and FAISS for efficient search...

API trace

Request

json

{"model": "gpt-4o", "messages": [{"role": "user", "content": "What is the main topic of the document?"}], "embedding": {"input": "document text"}}

Response

json

{"choices": [{"message": {"content": "The main topic is Retrieval-Augmented Generation (RAG)..."}}], "usage": {"total_tokens": 50}}

Extractresponse.choices[0].message.content

Variants

Streaming similarity search ›

Use when you want to see similarity scores alongside retrieved documents for better relevance understanding.

python

import os
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader

pdf_path = "example.pdf"
loader = PyPDFLoader(pdf_path)
docs = loader.load()

embeddings = OpenAIEmbeddings(api_key=os.environ["OPENAI_API_KEY"])
vector_store = FAISS.from_documents(docs, embeddings)

query = "Explain RAG in simple terms."
for doc in vector_store.similarity_search_with_score(query, k=3):
    print(f"Score: {doc[1]:.4f}, Content snippet: {doc[0].page_content[:150]}...")

Async PDF loading and embedding ›

Use in applications requiring concurrency or integration with async frameworks.

python

import os
import asyncio
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader

async def load_and_embed(pdf_path):
    loader = PyPDFLoader(pdf_path)
    docs = loader.load()
    embeddings = OpenAIEmbeddings(api_key=os.environ["OPENAI_API_KEY"])
    vector_store = FAISS.from_documents(docs, embeddings)
    return vector_store

async def main():
    vector_store = await load_and_embed("example.pdf")
    results = vector_store.similarity_search("What is RAG?", k=2)
    for doc in results:
        print(doc.page_content[:200])

asyncio.run(main())

Use Google Gemini embeddings instead of OpenAI ›

Use when you want to experiment with different embedding providers for cost or performance reasons.

python

import os
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain_openai import OpenAIEmbeddings

# Replace OpenAIEmbeddings with Gemini embeddings if available
# embeddings = GeminiEmbeddings(api_key=os.environ["GOOGLE_API_KEY"])

pdf_path = "example.pdf"
loader = PyPDFLoader(pdf_path)
docs = loader.load()

embeddings = OpenAIEmbeddings(api_key=os.environ["OPENAI_API_KEY"])
vector_store = FAISS.from_documents(docs, embeddings)

query = "Summarize the document."
results = vector_store.similarity_search(query, k=3)
for doc in results:
    print(doc.page_content[:200])

Performance

Latency~1-3 seconds for loading and embedding a 10-page PDF with OpenAI embeddings

Cost~$0.002 per 1,000 tokens embedded with OpenAI embeddings

Rate limitsOpenAI default tier: 350 RPM / 90,000 TPM

Split large PDFs into smaller chunks to avoid embedding huge texts at once.
Cache embeddings for static documents to reduce repeated API calls.
Use smaller embedding models if latency or cost is critical.

Approach	Latency	Cost/call	Best for
Standard load + FAISS	~2s	~$0.002 per 1k tokens	General RAG workflows
Streaming similarity search	~2-3s	~$0.002 per 1k tokens	When relevance scores matter
Async loading	~1-2s (concurrent)	~$0.002 per 1k tokens	High concurrency apps

✓

Quick tip

Always split PDFs into pages or chunks before embedding to improve retrieval accuracy in RAG.

⚠

Common mistake

Trying to embed entire PDFs as one document without splitting, which reduces retrieval precision.

Verified 2026-04 · gpt-4o, OpenAIEmbeddings, FAISS, PyPDFLoader

Verify ↗

Community Notes

No notes yetBe the first to share a version-specific fix or tip.