Getting Started with LangChain: Essential Concepts and Coding Examples

Written by Jens on 21 Aug 2025

Getting Started with LangChain: Essential Concepts and Coding Examples

LangChain is an open-source framework that helps developers build applications with large language models (LLMs).
It provides abstractions and integrations for working with LLMs like OpenAI, Anthropic, Cohere, and others.

What is LangChain?

At its core, LangChain makes it easier to connect:

LLMs (e.g. GPT-4, Claude, etc.)
Data sources (databases, PDFs, APIs, documents, etc.)
Chains (sequences of steps combining prompts, logic, and tools)
Agents (LLMs that decide which tools or actions to use)

Basic Examples

Here are some minimal Python examples:

1. Simple LLM Call

  from langchain_openai import OpenAI

  llm = OpenAI(model="gpt-3.5-turbo", temperature=0)
  response = llm.invoke("Explain LangChain in one sentence.")
  print(response)

2: Prompt Template + Chain

Sometimes you want to reuse a prompt in a structured way. Here we create a PromptTemplate, pass it to an LLMChain, and use it to generate consistent outputs.

  from langchain.prompts import PromptTemplate
  from langchain.chains import LLMChain
  from langchain_openai import OpenAI

  # Define a reusable prompt template
  prompt = PromptTemplate(
      input_variables=["topic", "audience"],
      template="Explain {topic} in simple terms for {audience}."
  )

  llm = OpenAI(model="gpt-3.5-turbo", temperature=0.5)
  chain = LLMChain(llm=llm, prompt=prompt)

  # Run the chain with different inputs
  print(chain.run({"topic": "quantum computing", "audience": "high school students"}))
  print(chain.run({"topic": "machine learning", "audience": "software engineers"}))

3: Document Q&A (Vectorstore)

Here’s how to let an LLM answer questions over your own data using embeddings + vector search.

  from langchain_openai import OpenAIEmbeddings, OpenAI
  from langchain_community.vectorstores import FAISS
  from langchain.chains import RetrievalQA

  # Step 1: Prepare documents (could also load from PDF, DB, API)
  docs = [
      "LangChain is a framework for building applications with LLMs.",
      "LangChain supports document Q&A, chatbots, and workflow automation.",
      "It integrates with vector databases like FAISS, Pinecone, Chroma."
  ]

  # Step 2: Build embeddings + vectorstore
  embeddings = OpenAIEmbeddings()
  docsearch = FAISS.from_texts(docs, embeddings)

  # Step 3: Build a Retrieval-based QA chain
  qa = RetrievalQA.from_chain_type(
      llm=OpenAI(model="gpt-3.5-turbo", temperature=0),
      retriever=docsearch.as_retriever(),
      return_source_documents=True
  )

  # Step 4: Ask questions
  query = "What can LangChain be used for?"
  result = qa.invoke(query)

  print("Answer:", result["result"])
  print("\nSources:")
  for doc in result["source_documents"]:
      print("-", doc.page_content)

4: Automation Workflow with an Agent

Agents allow an LLM to decide which tools to use dynamically (like calling APIs, running calculations, etc.). Here’s a simple workflow: the agent can answer general questions and do math.

  from langchain_openai import OpenAI
  from langchain.agents import initialize_agent, AgentType, load_tools

  # Step 1: Load tools (here we add a calculator tool)
  llm = OpenAI(model="gpt-3.5-turbo", temperature=0)
  tools = load_tools(["llm-math"], llm=llm)

  # Step 2: Initialize an agent
  agent = initialize_agent(
      tools=tools,
      llm=llm,
      agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
      verbose=True
  )

  # Step 3: Run queries
  print(agent.run("What's 17 times the square root of 2?"))
  print(agent.run("Summarize what LangChain is in 2 sentences."))

💡 Here the LLM chooses when to use the calculator tool vs. just generating text. In production, you can wire up tools like web search, APIs, databases, shell commands, etc.