Build a Retrieval Augmented Generation (RAG) App: Part 1

One of the most powerful applications enabled by LLMs is sophisticated question-answering (Q&A) chatbots. These are applications that can answer questions about specific source information. These applications use a technique known as Retrieval Augmented Generation, orRAG.

This is a multi-part tutorial:

Part 1 (this guide) introduces RAG and walks through a minimal implementation.
Part 2 extends the implementation to accommodate conversation-style interactions and multi-step retrieval processes.

This tutorial will show how to build a simple Q&A applicationover a text data source. Along the way we’ll go over a typical Q&Aarchitecture and highlight additional resources for more advanced Q&A techniques. We’ll also seehow LangSmith can help us trace and understand our application.LangSmith will become increasingly helpful as our application grows incomplexity.

If you're already familiar with basic retrieval, you might also be interested inthishigh-level overview of different retrieval techniques.

Note: Here we focus on Q&A for unstructured data. If you are interested for RAG over structured data, check out our tutorial on doingquestion/answering over SQL data.

Overview

A typical RAG application has two main components:

Indexing: a pipeline for ingesting data from a source and indexing it.This usually happens offline.

Retrieval and generation: the actual RAG chain, which takes the user query at run time and retrieves the relevant data from the index, then passes that to the model.

Note: the indexing portion of this tutorial will largely follow thesemantic search tutorial.

The most common full sequence from raw data to answer looks like:

Indexing

Load: First we need to load our data. This is done withDocument Loaders.
Split:Text splitters break largeDocuments into smaller chunks. This is useful both for indexing data and passing it into a model, as large chunks are harder to search over and won't fit in a model's finite context window.
Store: We need somewhere to store and index our splits, so that they can be searched over later. This is often done using aVectorStore andEmbeddings model.

index_diagram

Retrieval and generation

Retrieve: Given a user input, relevant splits are retrieved from storage using aRetriever.
Generate: AChatModel /LLM produces an answer using a prompt that includes both the question with the retrieved data

retrieval_diagram

Once we've indexed our data, we will useLangGraph as our orchestration framework to implement the retrieval and generation steps.

Setup

Jupyter Notebook

This and other tutorials are perhaps most conveniently run in aJupyter notebooks. Going through guides in an interactive environment is a great way to better understand them. Seehere for instructions on how to install.

Installation

This tutorial requires these langchain dependencies:

Pip
Conda

%pip install--quiet--upgrade langchain-text-splitters langchain-community langgraph

conda install langchain-text-splitters langchain-community langgraph -c conda-forge

For more details, see ourInstallation guide.

LangSmith

Many of the applications you build with LangChain will contain multiple steps with multiple invocations of LLM calls.As these applications get more complex, it becomes crucial to be able to inspect what exactly is going on inside your chain or agent.The best way to do this is withLangSmith.

After you sign up at the link above, make sure to set your environment variables to start logging traces:

export LANGSMITH_TRACING="true"
export LANGSMITH_API_KEY="..."

Or, if in a notebook, you can set them with:

import getpass
import os

os.environ["LANGSMITH_TRACING"]="true"
os.environ["LANGSMITH_API_KEY"]= getpass.getpass()

Components

We will need to select three components from LangChain's suite of integrations.

Selectchat model:

pip install -qU "langchain[google-genai]"

import getpass
import os

ifnot os.environ.get("GOOGLE_API_KEY"):
  os.environ["GOOGLE_API_KEY"]= getpass.getpass("Enter API key for Google Gemini: ")

from langchain.chat_modelsimport init_chat_model

llm= init_chat_model("gemini-2.0-flash", model_provider="google_genai")

Selectembeddings model:

pip install -qU langchain-openai

import getpass
import os

ifnot os.environ.get("OPENAI_API_KEY"):
  os.environ["OPENAI_API_KEY"]= getpass.getpass("Enter API key for OpenAI: ")

from langchain_openaiimport OpenAIEmbeddings

embeddings= OpenAIEmbeddings(model="text-embedding-3-large")

Selectvector store:

pip install -qU langchain-core

from langchain_core.vectorstoresimport InMemoryVectorStore

vector_store= InMemoryVectorStore(embeddings)

Preview

In this guide we’ll build an app that answers questions about the website's content. The specific website we will use is theLLM Powered AutonomousAgents blog postby Lilian Weng, which allows us to ask questions about the contents ofthe post.

We can create a simple indexing pipeline and RAG chain to do this in ~50lines of code.

import bs4
from langchainimport hub
from langchain_community.document_loadersimport WebBaseLoader
from langchain_core.documentsimport Document
from langchain_text_splittersimport RecursiveCharacterTextSplitter
from langgraph.graphimport START, StateGraph
from typing_extensionsimport List, TypedDict

# Load and chunk contents of the blog
loader= WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs=dict(
        parse_only=bs4.SoupStrainer(
            class_=("post-content","post-title","post-header")
)
),
)
docs= loader.load()

text_splitter= RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
all_splits= text_splitter.split_documents(docs)

# Index chunks
_= vector_store.add_documents(documents=all_splits)

# Define prompt for question-answering
# N.B. for non-US LangSmith endpoints, you may need to specify
# api_url="https://api.smith.langchain.com" in hub.pull.
prompt= hub.pull("rlm/rag-prompt")


# Define state for application
classState(TypedDict):
    question:str
    context: List[Document]
    answer:str


# Define application steps
defretrieve(state: State):
    retrieved_docs= vector_store.similarity_search(state["question"])
return{"context": retrieved_docs}


defgenerate(state: State):
    docs_content="\n\n".join(doc.page_contentfor docin state["context"])
    messages= prompt.invoke({"question": state["question"],"context": docs_content})
    response= llm.invoke(messages)
return{"answer": response.content}


# Compile application and test
graph_builder= StateGraph(State).add_sequence([retrieve, generate])
graph_builder.add_edge(START,"retrieve")
graph= graph_builder.compile()

API Reference:hub |WebBaseLoader |Document |RecursiveCharacterTextSplitter |StateGraph

response= graph.invoke({"question":"What is Task Decomposition?"})
print(response["answer"])

Task Decomposition is the process of breaking down a complicated task into smaller, manageable steps to facilitate easier execution and understanding. Techniques like Chain of Thought (CoT) and Tree of Thoughts (ToT) guide models to think step-by-step, allowing them to explore multiple reasoning possibilities. This method enhances performance on complex tasks and provides insight into the model's thinking process.

Check out theLangSmithtrace.

Detailed walkthrough

Let’s go through the above code step-by-step to really understand what’sgoing on.

1. Indexing

note

This section is an abbreviated version of the content in thesemantic search tutorial.If you're comfortable withdocument loaders,embeddings, andvector stores,feel free to skip to the next section onretrieval and generation.

Loading documents

We need to first load the blog post contents. We can useDocumentLoadersfor this, which are objects that load in data from a source and return alist ofDocumentobjects.

In this case we’ll use theWebBaseLoader,which usesurllib to load HTML from web URLs andBeautifulSoup toparse it to text. We can customize the HTML -> text parsing by passingin parameters into theBeautifulSoup parser viabs_kwargs (seeBeautifulSoupdocs).In this case only HTML tags with class “post-content”, “post-title”, or“post-header” are relevant, so we’ll remove all others.

import bs4
from langchain_community.document_loadersimport WebBaseLoader

# Only keep post title, headers, and content from the full HTML.
bs4_strainer= bs4.SoupStrainer(class_=("post-title","post-header","post-content"))
loader= WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs={"parse_only": bs4_strainer},
)
docs= loader.load()

assertlen(docs)==1
print(f"Total characters:{len(docs[0].page_content)}")

API Reference:WebBaseLoader

Total characters: 43131

print(docs[0].page_content[:500])



      LLM Powered Autonomous Agents

Date: June 23, 2023  |  Estimated Reading Time: 31 min  |  Author: Lilian Weng


Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.
Agent System Overview#
In

Go deeper

DocumentLoader: Object that loads data from a source as list ofDocuments.

Docs:Detailed documentation on how to useDocumentLoaders.
Integrations: 160+integrations to choose from.
Interface:API reference for the base interface.

Splitting documents

Our loaded document is over 42k characters which is too long to fitinto the context window of many models. Even for those models that couldfit the full post in their context window, models can struggle to findinformation in very long inputs.

To handle this we’ll split theDocument into chunks for embedding andvector storage. This should help us retrieve only the most relevant partsof the blog post at run time.

As in thesemantic search tutorial, we use aRecursiveCharacterTextSplitter,which will recursively split the document using common separators likenew lines until each chunk is the appropriate size. This is therecommended text splitter for generic text use cases.

from langchain_text_splittersimport RecursiveCharacterTextSplitter

text_splitter= RecursiveCharacterTextSplitter(
    chunk_size=1000,# chunk size (characters)
    chunk_overlap=200,# chunk overlap (characters)
    add_start_index=True,# track index in original document
)
all_splits= text_splitter.split_documents(docs)

print(f"Split blog post into{len(all_splits)} sub-documents.")

API Reference:RecursiveCharacterTextSplitter

Split blog post into 66 sub-documents.

Go deeper

TextSplitter: Object that splits a list ofDocuments into smallerchunks. Subclass ofDocumentTransformers.

Learn more about splitting text using different methods by reading thehow-to docs
Code (py or js)
Scientific papers
Interface: API reference for the base interface.

DocumentTransformer: Object that performs a transformation on a listofDocument objects.

Docs: Detailed documentation on how to useDocumentTransformers
Integrations
Interface: API reference for the base interface.

Storing documents

Now we need to index our 66 text chunks so that we can search over themat runtime. Following thesemantic search tutorial,our approach is toembed the contents of each document split and insert these embeddingsinto avector store. Given an input query, we can then usevector search to retrieve relevant documents.

We can embed and store all of our document splits in a single commandusing the vector store and embeddings model selected at thestart of the tutorial.

document_ids= vector_store.add_documents(documents=all_splits)

print(document_ids[:3])

['07c18af6-ad58-479a-bfb1-d508033f9c64', '9000bf8e-1993-446f-8d4d-f4e507ba4b8f', 'ba3b5d14-bed9-4f5f-88be-44c88aedc2e6']

Go deeper

Embeddings: Wrapper around a text embedding model, used for convertingtext to embeddings.

Docs: Detailed documentation on how to use embeddings.
Integrations: 30+ integrations to choose from.
Interface: API reference for the base interface.

VectorStore: Wrapper around a vector database, used for storing andquerying embeddings.

Docs: Detailed documentation on how to use vector stores.
Integrations: 40+ integrations to choose from.
Interface: API reference for the base interface.

This completes theIndexing portion of the pipeline. At this pointwe have a query-able vector store containing the chunked contents of ourblog post. Given a user question, we should ideally be able to returnthe snippets of the blog post that answer the question.

2. Retrieval and Generation

Now let’s write the actual application logic. We want to create a simpleapplication that takes a user question, searches for documents relevantto that question, passes the retrieved documents and initial question toa model, and returns an answer.

For generation, we will use the chat model selected at thestart of the tutorial.

We’ll use a prompt for RAG that is checked into the LangChain prompt hub(here).

from langchainimport hub

# N.B. for non-US LangSmith endpoints, you may need to specify
# api_url="https://api.smith.langchain.com" in hub.pull.
prompt= hub.pull("rlm/rag-prompt")

example_messages= prompt.invoke(
{"context":"(context goes here)","question":"(question goes here)"}
).to_messages()

assertlen(example_messages)==1
print(example_messages[0].content)

API Reference:hub

You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, just say that you don't know. Use three sentences maximum and keep the answer concise.
Question: (question goes here)
Context: (context goes here)
Answer:

We'll useLangGraph to tie together the retrieval and generation steps into a single application. This will bring a number of benefits:

We can define our application logic once and automatically support multiple invocation modes, including streaming, async, and batched calls.
We get streamlined deployments viaLangGraph Platform.
LangSmith will automatically trace the steps of our application together.
We can easily add key features to our application, includingpersistence andhuman-in-the-loop approval, with minimal code changes.

To use LangGraph, we need to define three things:

The state of our application;
The nodes of our application (i.e., application steps);
The "control flow" of our application (e.g., the ordering of the steps).

State:

Thestate of our application controls what data is input to the application, transferred between steps, and output by the application. It is typically aTypedDict, but can also be aPydantic BaseModel.

For a simple RAG application, we can just keep track of the input question, retrieved context, and generated answer:

from langchain_core.documentsimport Document
from typing_extensionsimport List, TypedDict


classState(TypedDict):
    question:str
    context: List[Document]
    answer:str

API Reference:Document

Nodes (application steps)

Let's start with a simple sequence of two steps: retrieval and generation.

defretrieve(state: State):
    retrieved_docs= vector_store.similarity_search(state["question"])
return{"context": retrieved_docs}


defgenerate(state: State):
    docs_content="\n\n".join(doc.page_contentfor docin state["context"])
    messages= prompt.invoke({"question": state["question"],"context": docs_content})
    response= llm.invoke(messages)
return{"answer": response.content}

Our retrieval step simply runs a similarity search using the input question, and the generation step formats the retrieved context and original question into a prompt for the chat model.

Control flow

Finally, we compile our application into a singlegraph object. In this case, we are just connecting the retrieval and generation steps into a single sequence.

from langgraph.graphimport START, StateGraph

graph_builder= StateGraph(State).add_sequence([retrieve, generate])
graph_builder.add_edge(START,"retrieve")
graph= graph_builder.compile()

API Reference:StateGraph

LangGraph also comes with built-in utilities for visualizing the control flow of your application:

from IPython.displayimport Image, display

display(Image(graph.get_graph().draw_mermaid_png()))

Do I need to use LangGraph?

LangGraph is not required to build a RAG application. Indeed, we can implement the same application logic through invocations of the individual components:

question="..."

retrieved_docs= vector_store.similarity_search(question)
docs_content="\n\n".join(doc.page_contentfor docin retrieved_docs)
prompt= prompt.invoke({"question": question,"context": docs_content})
answer= llm.invoke(prompt)

The benefits of LangGraph include:

Support for multiple invocation modes: this logic would need to be rewritten if we wanted to stream output tokens, or stream the results of individual steps;
Automatic support for tracing viaLangSmith and deployments viaLangGraph Platform;
Support for persistence, human-in-the-loop, and other features.

Many use-cases demand RAG in a conversational experience, such that a user can receive context-informed answers via a stateful conversation. As we will see inPart 2 of the tutorial, LangGraph's management and persistence of state simplifies these applications enormously.

Usage

Let's test our application! LangGraph supports multiple invocation modes, including sync, async, and streaming.

Invoke:

result= graph.invoke({"question":"What is Task Decomposition?"})

print(f"Context:{result['context']}\n\n")
print(f"Answer:{result['answer']}")

Context: [Document(id='a42dc78b-8f76-472a-9e25-180508af74f3', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 1585}, page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\nComponent One: Planning#\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\nTask Decomposition#\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.'), Document(id='c0e45887-d0b0-483d-821a-bb5d8316d51d', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 2192}, page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\nTask decomposition can be done (1) by LLM with simple prompting like "Steps for XYZ.\\n1.", "What are the subgoals for achieving XYZ?", (2) by using task-specific instructions; e.g. "Write a story outline." for writing a novel, or (3) with human inputs.'), Document(id='4cc7f318-35f5-440f-a4a4-145b5f0b918d', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 29630}, page_content='Resources:\n1. Internet access for searches and information gathering.\n2. Long Term memory management.\n3. GPT-3.5 powered Agents for delegation of simple tasks.\n4. File output.\n\nPerformance Evaluation:\n1. Continuously review and analyze your actions to ensure you are performing to the best of your abilities.\n2. Constructively self-criticize your big-picture behavior constantly.\n3. Reflect on past decisions and strategies to refine your approach.\n4. Every command has a cost, so be smart and efficient. Aim to complete tasks in the least number of steps.'), Document(id='f621ade4-9b0d-471f-a522-44eb5feeba0c', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 19373}, page_content="(3) Task execution: Expert models execute on the specific tasks and log results.\nInstruction:\n\nWith the input and the inference results, the AI assistant needs to describe the process and results. The previous stages can be formed as - User Input: {{ User Input }}, Task Planning: {{ Tasks }}, Model Selection: {{ Model Assignment }}, Task Execution: {{ Predictions }}. You must first answer the user's request in a straightforward manner. Then describe the task process and show your analysis and model inference results to the user in the first person. If inference results contain a file path, must tell the user the complete file path.")]


Answer: Task decomposition is a technique used to break down complex tasks into smaller, manageable steps, allowing for more efficient problem-solving. This can be achieved through methods like chain of thought prompting or the tree of thoughts approach, which explores multiple reasoning possibilities at each step. It can be initiated through simple prompts, task-specific instructions, or human inputs.

Stream steps:

for stepin graph.stream(
{"question":"What is Task Decomposition?"}, stream_mode="updates"
):
print(f"{step}\n\n----------------\n")

{'retrieve': {'context': [Document(id='a42dc78b-8f76-472a-9e25-180508af74f3', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 1585}, page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\nComponent One: Planning#\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\nTask Decomposition#\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.'), Document(id='c0e45887-d0b0-483d-821a-bb5d8316d51d', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 2192}, page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\nTask decomposition can be done (1) by LLM with simple prompting like "Steps for XYZ.\\n1.", "What are the subgoals for achieving XYZ?", (2) by using task-specific instructions; e.g. "Write a story outline." for writing a novel, or (3) with human inputs.'), Document(id='4cc7f318-35f5-440f-a4a4-145b5f0b918d', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 29630}, page_content='Resources:\n1. Internet access for searches and information gathering.\n2. Long Term memory management.\n3. GPT-3.5 powered Agents for delegation of simple tasks.\n4. File output.\n\nPerformance Evaluation:\n1. Continuously review and analyze your actions to ensure you are performing to the best of your abilities.\n2. Constructively self-criticize your big-picture behavior constantly.\n3. Reflect on past decisions and strategies to refine your approach.\n4. Every command has a cost, so be smart and efficient. Aim to complete tasks in the least number of steps.'), Document(id='f621ade4-9b0d-471f-a522-44eb5feeba0c', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 19373}, page_content="(3) Task execution: Expert models execute on the specific tasks and log results.\nInstruction:\n\nWith the input and the inference results, the AI assistant needs to describe the process and results. The previous stages can be formed as - User Input: {{ User Input }}, Task Planning: {{ Tasks }}, Model Selection: {{ Model Assignment }}, Task Execution: {{ Predictions }}. You must first answer the user's request in a straightforward manner. Then describe the task process and show your analysis and model inference results to the user in the first person. If inference results contain a file path, must tell the user the complete file path.")]}}

----------------

{'generate': {'answer': 'Task decomposition is the process of breaking down a complex task into smaller, more manageable steps. This technique, often enhanced by methods like Chain of Thought (CoT) or Tree of Thoughts, allows models to reason through tasks systematically and improves performance by clarifying the thought process. It can be achieved through simple prompts, task-specific instructions, or human inputs.'}}

----------------

Streamtokens:

for message, metadatain graph.stream(
{"question":"What is Task Decomposition?"}, stream_mode="messages"
):
print(message.content, end="|")

|Task| decomposition| is| the| process| of| breaking| down| complex| tasks| into| smaller|,| more| manageable| steps|.| It| can| be| achieved| through| techniques| like| Chain| of| Thought| (|Co|T|)| prompting|,| which| encourages| the| model| to| think| step| by| step|,| or| through| more| structured| methods| like| the| Tree| of| Thoughts|.| This| approach| not| only| simplifies| task| execution| but| also| provides| insights| into| the| model|'s| reasoning| process|.||

tip

For async invocations, use:

result=await graph.ainvoke(...)

and

asyncfor stepin graph.astream(...):

Returning sources

Note that by storing the retrieved context in the state of the graph, we recover sources for the model's generated answer in the"context" field of the state. Seethis guide on returning sources for more detail.

Go deeper

Chat models take in a sequence of messages and return a message.

Docs
Integrations: 25+ integrations to choose from.
Interface: API reference for the base interface.

Customizing the prompt

As shown above, we can load prompts (e.g.,this RAGprompt) from the prompthub. The prompt can also be easily customized. For example:

from langchain_core.promptsimport PromptTemplate

template="""Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
Use three sentences maximum and keep the answer as concise as possible.
Always say "thanks for asking!" at the end of the answer.

{context}

Question: {question}

Helpful Answer:"""
custom_rag_prompt= PromptTemplate.from_template(template)

API Reference:PromptTemplate

Query analysis

So far, we are executing the retrieval using the raw input query. However, there are some advantages to allowing a model to generate the query for retrieval purposes. For example:

In addition to semantic search, we can build in structured filters (e.g., "Find documents since the year 2020.");
The model can rewrite user queries, which may be multifaceted or include irrelevant language, into more effective search queries.

Query analysis employs models to transform or construct optimized search queries from raw user input. We can easily incorporate a query analysis step into our application. For illustrative purposes, let's add some metadata to the documents in our vector store. We will add some (contrived) sections to the document which we can filter on later.

total_documents=len(all_splits)
third= total_documents//3

for i, documentinenumerate(all_splits):
if i< third:
        document.metadata["section"]="beginning"
elif i<2* third:
        document.metadata["section"]="middle"
else:
        document.metadata["section"]="end"


all_splits[0].metadata

{'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/',
 'start_index': 8,
 'section': 'beginning'}

We will need to update the documents in our vector store. We will use a simpleInMemoryVectorStore for this, as we will use some of its specific features (i.e., metadata filtering). Refer to the vector storeintegration documentation for relevant features of your chosen vector store.

from langchain_core.vectorstoresimport InMemoryVectorStore

vector_store= InMemoryVectorStore(embeddings)
_= vector_store.add_documents(all_splits)

API Reference:InMemoryVectorStore

Let's next define a schema for our search query. We will usestructured output for this purpose. Here we define a query as containing a string query and a document section (either "beginning", "middle", or "end"), but this can be defined however you like.

from typingimport Literal

from typing_extensionsimport Annotated


classSearch(TypedDict):
"""Search query."""

    query: Annotated[str,...,"Search query to run."]
    section: Annotated[
        Literal["beginning","middle","end"],
...,
"Section to query.",
]

Finally, we add a step to our LangGraph application to generate a query from the user's raw input:

classState(TypedDict):
    question:str
    query: Search
    context: List[Document]
    answer:str


defanalyze_query(state: State):
    structured_llm= llm.with_structured_output(Search)
    query= structured_llm.invoke(state["question"])
return{"query": query}


defretrieve(state: State):
    query= state["query"]
    retrieved_docs= vector_store.similarity_search(
        query["query"],
filter=lambda doc: doc.metadata.get("section")== query["section"],
)
return{"context": retrieved_docs}


defgenerate(state: State):
    docs_content="\n\n".join(doc.page_contentfor docin state["context"])
    messages= prompt.invoke({"question": state["question"],"context": docs_content})
    response= llm.invoke(messages)
return{"answer": response.content}


graph_builder= StateGraph(State).add_sequence([analyze_query, retrieve, generate])
graph_builder.add_edge(START,"analyze_query")
graph= graph_builder.compile()

Full Code:

from typingimport Literal

import bs4
from langchainimport hub
from langchain_community.document_loadersimport WebBaseLoader
from langchain_core.documentsimport Document
from langchain_core.vectorstoresimport InMemoryVectorStore
from langchain_text_splittersimport RecursiveCharacterTextSplitter
from langgraph.graphimport START, StateGraph
from typing_extensionsimport Annotated, List, TypedDict

# Load and chunk contents of the blog
loader= WebBaseLoader(
    web_paths=("https://lilianweng.github.io/posts/2023-06-23-agent/",),
    bs_kwargs=dict(
        parse_only=bs4.SoupStrainer(
            class_=("post-content","post-title","post-header")
)
),
)
docs= loader.load()

text_splitter= RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
all_splits= text_splitter.split_documents(docs)


# Update metadata (illustration purposes)
total_documents=len(all_splits)
third= total_documents//3

for i, documentinenumerate(all_splits):
if i< third:
        document.metadata["section"]="beginning"
elif i<2* third:
        document.metadata["section"]="middle"
else:
        document.metadata["section"]="end"


# Index chunks
vector_store= InMemoryVectorStore(embeddings)
_= vector_store.add_documents(all_splits)


# Define schema for search
classSearch(TypedDict):
"""Search query."""

    query: Annotated[str,...,"Search query to run."]
    section: Annotated[
        Literal["beginning","middle","end"],
...,
"Section to query.",
]

# Define prompt for question-answering
prompt= hub.pull("rlm/rag-prompt")


# Define state for application
classState(TypedDict):
    question:str
    query: Search
    context: List[Document]
    answer:str


defanalyze_query(state: State):
    structured_llm= llm.with_structured_output(Search)
    query= structured_llm.invoke(state["question"])
return{"query": query}


defretrieve(state: State):
    query= state["query"]
    retrieved_docs= vector_store.similarity_search(
        query["query"],
filter=lambda doc: doc.metadata.get("section")== query["section"],
)
return{"context": retrieved_docs}


defgenerate(state: State):
    docs_content="\n\n".join(doc.page_contentfor docin state["context"])
    messages= prompt.invoke({"question": state["question"],"context": docs_content})
    response= llm.invoke(messages)
return{"answer": response.content}


graph_builder= StateGraph(State).add_sequence([analyze_query, retrieve, generate])
graph_builder.add_edge(START,"analyze_query")
graph= graph_builder.compile()

display(Image(graph.get_graph().draw_mermaid_png()))

We can test our implementation by specifically asking for context from the end of the post. Note that the model includes different information in its answer.

for stepin graph.stream(
{"question":"What does the end of the post say about Task Decomposition?"},
    stream_mode="updates",
):
print(f"{step}\n\n----------------\n")

{'analyze_query': {'query': {'query': 'Task Decomposition', 'section': 'end'}}}

----------------

{'retrieve': {'context': [Document(id='d6cef137-e1e8-4ddc-91dc-b62bd33c6020', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 39221, 'section': 'end'}, page_content='Finite context length: The restricted context capacity limits the inclusion of historical information, detailed instructions, API call context, and responses. The design of the system has to work with this limited communication bandwidth, while mechanisms like self-reflection to learn from past mistakes would benefit a lot from long or infinite context windows. Although vector stores and retrieval can provide access to a larger knowledge pool, their representation power is not as powerful as full attention.\n\n\nChallenges in long-term planning and task decomposition: Planning over a lengthy history and effectively exploring the solution space remain challenging. LLMs struggle to adjust plans when faced with unexpected errors, making them less robust compared to humans who learn from trial and error.'), Document(id='d1834ae1-eb6a-43d7-a023-08dfa5028799', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 39086, 'section': 'end'}, page_content='}\n]\nChallenges#\nAfter going through key ideas and demos of building LLM-centered agents, I start to see a couple common limitations:'), Document(id='ca7f06e4-2c2e-4788-9a81-2418d82213d9', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 32942, 'section': 'end'}, page_content='}\n]\nThen after these clarification, the agent moved into the code writing mode with a different system message.\nSystem message:'), Document(id='1fcc2736-30f4-4ef6-90f2-c64af92118cb', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'start_index': 35127, 'section': 'end'}, page_content='"content": "You will get instructions for code to write.\\nYou will write a very long answer. Make sure that every detail of the architecture is, in the end, implemented as code.\\nMake sure that every detail of the architecture is, in the end, implemented as code.\\n\\nThink step by step and reason yourself to the right decisions to make sure we get it right.\\nYou will first lay out the names of the core classes, functions, methods that will be necessary, as well as a quick comment on their purpose.\\n\\nThen you will output the content of each file including ALL code.\\nEach file must strictly follow a markdown code block format, where the following tokens must be replaced such that\\nFILENAME is the lowercase file name including the file extension,\\nLANG is the markup code block language for the code\'s language, and CODE is the code:\\n\\nFILENAME\\n\`\`\`LANG\\nCODE\\n\`\`\`\\n\\nYou will start with the \\"entrypoint\\" file, then go to the ones that are imported by that file, and so on.\\nPlease')]}}

----------------

{'generate': {'answer': 'The end of the post highlights that task decomposition faces challenges in long-term planning and adapting to unexpected errors. LLMs struggle with adjusting their plans, making them less robust compared to humans who learn from trial and error. This indicates a limitation in effectively exploring the solution space and handling complex tasks.'}}

----------------

In both the streamed steps and theLangSmith trace, we can now observe the structured query that was fed into the retrieval step.

Query Analysis is a rich problem with a wide range of approaches. Refer to thehow-to guides for more examples.

Next steps

We've covered the steps to build a basic Q&A app over data:

Loading data with aDocument Loader
Chunking the indexed data with aText Splitter to make it more easily usable by a model
Embedding the data and storing the data in avectorstore
Retrieving the previously stored chunks in response to incoming questions
Generating an answer using the retrieved chunks as context.

InPart 2 of the tutorial, we will extend the implementation here to accommodate conversation-style interactions and multi-step retrieval processes.

Movatterモバイル変換

Overview​

Indexing​

Retrieval and generation​

Setup​

Jupyter Notebook​

Installation​

LangSmith​

Components​

Preview​

Detailed walkthrough​

1. Indexing​

Loading documents​

Go deeper​

Splitting documents​

Go deeper​

Storing documents​

Go deeper​

2. Retrieval and Generation​

State:​

Nodes (application steps)​

Control flow​

Usage​

Returning sources​

Go deeper​

Query analysis​

Next steps​

Overview

Indexing

Retrieval and generation

Setup

Jupyter Notebook

Installation

LangSmith

Components

Preview

Detailed walkthrough

1. Indexing

Loading documents

Go deeper

Splitting documents

Go deeper

Storing documents

Go deeper

2. Retrieval and Generation

State:

Nodes (application steps)

Control flow

Usage

Returning sources

Go deeper

Query analysis

Next steps