Python LangChain

Installing LangChain Packages with uv

• LangChain uses separate packages for core primitives and provider-specific integrations like OpenAI.
• We need langchain-core for ChatPromptTemplate, chaining (|), and StrOutputParser.
• We need langchain-openai to use ChatOpenAI with the OpenAI chat models like gpt-4o.
• Install required packages using uv in your project environment as shown below.
• Commands Used:
  • uv add langchain-core
  • uv add langchain-openai #if open ai used
  • uv add langchain-google-genai #if google gen ai used

Understanding the LangChain chain = prompt | llm | parser

• The | operator creates a sequential pipeline: output of left side becomes input of right side.
• First, the prompt turns {"input": value} into a full prompt object for the chat model.
• Next, the llm (either ChatOpenAI or ChatGoogleGenerativeAI) takes that prompt and generates a raw AI response.
• Finally, StrOutputParser converts the raw response into a plain Python string suitable for printing.
• This chain abstraction keeps the chat flow clean, readable, and easy to swap between OpenAI and Gemini.

ChatPromptTemplate:

• ChatPromptTemplate is used to define a prompt template instead of passing raw strings every time.
• In the code, ChatPromptTemplate.from_template("{input}") creates a template with a single placeholder called {input}.
• We need this so that whatever the user types can be injected cleanly into the prompt each time.
• During execution, the chain replaces {input} with the current value entered by the user in the loop.
• This keeps the prompt logic in one place and makes the chain reusable for every user message.

OpenAI Chat Chain:

• This chain uses OpenAI’s gpt-4o model wrapped by LangChain’s ChatOpenAI class.
• We need this chain to take user input, send it to the OpenAI model, and convert the model’s reply into a simple Python string.
• The chain is built as prompt | llm | parser, meaning: template → model → output parsing.
• Inside the loop, each user value is passed as {"input": value`} into chain.invoke(...), and the answer is printed.
• The loop runs until the user types exit or quit, giving a simple console-style chatbot.

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser

prompt = ChatPromptTemplate.from_template("{input}")

llm = ChatOpenAI(model="gpt-4o")

parser = StrOutputParser()

while True:
    value = input("you: ")
    if value.lower() in ["exit", "quit"]:
        break

    chain = prompt | llm | parser

    answer = chain.invoke({"input": value})
    print("AI: " + answer)

Gemini Chat Chain:

• This chain uses Google’s Gemini model wrapped by LangChain’s ChatGoogleGenerativeAI class.
• We need this version when we want to switch the backend LLM from OpenAI to Gemini with minimal code changes.
• The structure is the same: prompt | llm | parser, only the llm object and model name ("gemini-2.0-flash") are different.
• The same input loop is reused: user types a message, it flows through the chain, and the AI’s reply is printed.
• This pattern shows how changing providers is mostly a matter of swapping the LLM class and model name.

from langchain_core.prompts import ChatPromptTemplate
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_core.output_parsers import StrOutputParser

prompt = ChatPromptTemplate.from_template("{input}")

llm = ChatGoogleGenerativeAI(
    model="gemini-2.0-flash"
)

parser = StrOutputParser()

while True:
    value = input("you: ")
    if value.lower() in ["exit", "quit"]:
        break

    chain = prompt | llm | parser

    answer = chain.invoke({"input": value})
    print("AI:", answer)