3. Chat Model Basic Conversation

Structure conversations using System, Human, and AI messages

In this section, you will learn how to structure multi-turn conversations using LangChain’s message schemas.

Objectives

1. Structure conversations with SystemMessage, HumanMessage, and AIMessage.
2. Prime model behavior with System messages.
3. Simulate multi-turn dialogue memory.

Understanding Chat Message Roles

When building multi-turn conversations, LLMs require messages to be structured with specific roles. LangChain provides specialized API classes under langchain_core.messages to handle these roles:

• SystemMessage: Sets the persona, behavior, constraints, and instructions for the AI model (e.g., “You are a helpful French translation assistant”).
• HumanMessage: Represents the queries or prompts sent directly by the user.
• AIMessage: Represents the responses returned by the AI model. In multi-turn chat logs, previous AI responses are passed back to the model as AIMessage instances to simulate context memory.

Conversation Structure

Goal

Instruct the model to act as a math solver and provide the history of a previous calculation.

Sample Input

Sequential messages:

1. SystemMessage(content="Solve the following math problems")
2. HumanMessage(content="What is 81 divided by 9?")
3. AIMessage(content="81 divided by 9 is 9.")
4. HumanMessage(content="What is 10 times 5?")

Sample Output

"10 times 5 is 50."

Plan

1. Import SystemMessage, HumanMessage, and AIMessage.
2. Package messages in a list representing the chat conversation history.
3. Pass the list to model.invoke() and print the response.

Code Implementation

1. Setup and Invoke

from dotenv import load_dotenv
from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
from langchain.chat_models import init_chat_model

load_dotenv()

# Initialize model using core abstractions (Google Gemini)
model = init_chat_model("gemini-2.5-flash", model_provider="google_genai")

# Package conversation history
messages = [
    SystemMessage(content="Solve the following math problems"),
    HumanMessage(content="What is 81 divided by 9?"),
    AIMessage(content="81 divided by 9 is 9."),
    HumanMessage(content="What is 10 times 5?"),
]

result = model.invoke(messages)
print(f"Answer from AI: {result.content}")

Exercise: Translation Memory 🌍

Goal

Build a conversation history list where the user instructs the model to translate words to French, translates one word, and then queries a second word.

Sample Input

1. System message: “Translate words to French.”
2. Human message: “Cat” -> AI: “Chat”
3. Human message: “Dog”

Sample Output

"Chien"

Plan

1. Form a conversation list using SystemMessage, HumanMessage, and AIMessage.
2. Call model.invoke() with the list using a Groq-provided model (llama-3.3-70b-versatile) initialized via core abstractions and print the output.

Solution

from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
from langchain.chat_models import init_chat_model
from dotenv import load_dotenv

load_dotenv()

# Initialize model using core abstractions (Groq Llama model)
model = init_chat_model("llama-3.3-70b-versatile", model_provider="groq")

messages = [
    SystemMessage(content="You are a French translator."),
    HumanMessage(content="Cat"),
    AIMessage(content="Chat"),
    HumanMessage(content="Dog")
]
res = model.invoke(messages)
print(res.content)

Practice & Exercises

To practice, open the interactive notebook:

Practice & Exercises

Practice structuring conversational message flows.

💻 VS Code | 🚀 Colab | 📥 Download