Skip to main content

LangChain Architecture Fundamentals

LangChain provides a unified interface for composing Large Language Models with external data, tools, and memory modules.

Core Abstractions

LangChain standardizes model interactions across providers (OpenAI, Google Gemini, Anthropic, Ollama, HuggingFace) into modular primitives.

Key Modules​

1. Model Interfaces​

LangChain separates model primitives into two categories:

  • ChatModels: Take a sequence of structured messages (SystemMessage, HumanMessage, AIMessage) and return a ChatResult.
  • LLMs: Take a plain text string and return a plain text completion string.
from langchain_core.messages import SystemMessage, HumanMessage
from langchain_openai import ChatOpenAI

model = ChatOpenAI(model="gpt-4o", temperature=0.2)

messages = [
    SystemMessage(content="You are an expert AI systems architect."),
    HumanMessage(content="Explain the difference between Bi-Encoders and Cross-Encoders.")
]

response = model.invoke(messages)
print(response.content)

2. Prompt Templates​

PromptTemplates dynamicize static text prompts with input variables:

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are an expert code reviewer specializing in {language}."),
    ("user", "Review the following code snippet:\n{code}")
])

formatted_messages = prompt.format_messages(
    language="Python",
    code="def add(a, b): return a + b"
)

3. Output Parsers​

Transform raw unstructured text responses from language models into structured formats (JSON, Pydantic objects, CSVs):

from pydantic import BaseModel, Field
from langchain_core.output_parsers import PydanticOutputParser

class TechnicalSummary(BaseModel):
    key_findings: list[str] = Field(description="List of core insights")
    complexity_score: int = Field(description="Score from 1 to 10")

parser = PydanticOutputParser(pydantic_object=TechnicalSummary)

4. Tools & Tool Calling​

Equip models with external execution capabilities (web search, database queries, calculators):

from langchain_core.tools import tool

@tool
def calculate_matrix_norm(vector: list[float]) -> float:
    """Calculates the Euclidean norm of a floating point vector."""
    return sum(x**2 for x in vector) ** 0.5

model_with_tools = model.bind_tools([calculate_matrix_norm])