🧩 Full Code:

contextualize_question_prompt = ChatPromptTemplate.from_messages([
    ("system", (
        "Given a conversation history and the most recent user query, rewrite the query as a standalone question "
        "that makes sense without relying on the previous context. Do not provide an answer—only reformulate the "
        "question if necessary; otherwise, return it unchanged."
    )),
    MessagesPlaceholder("chat_history"),
    ("human", "{input}"),
])

⚙️ Purpose (in simple terms)

This prompt template is designed to rewrite a user’s query so that it’s self-contained — meaning the question can stand on its own, without needing the earlier chat context.

It’s often used in retrieval-augmented generation (RAG) or chat memory pipelines in LangChain to make the user’s query more meaningful before searching documents or giving responses.

🧠 Component-by-Component Explanation

1. ChatPromptTemplate

From: langchain_core.prompts.ChatPromptTemplate

This is a LangChain class used to define structured prompts for chat-based models (like GPT).
It helps combine multiple message types — system, user (human), AI, etc. — into one prompt sequence that an LLM can understand.

Why use it:
It separates the roles and content of each message (like system instructions, previous messages, user input) — making the prompt modular and reusable.

2. .from_messages([...])

This creates the ChatPromptTemplate from a list of message tuples or message-like objects.
Each element in the list defines a role and the content for that message.

The roles can be:

• "system" → sets the rules or behavior of the AI.

• "human" → represents user input.

• "ai" → represents model responses (if used).

• MessagesPlaceholder(...) → placeholder for dynamically inserted messages.

3. ("system", "...")

This defines the system message.

🗣️ The model reads it like a “rulebook” before responding.

In your code:

("system", (
    "Given a conversation history and the most recent user query, rewrite the query as a standalone question "
    "that makes sense without relying on the previous context. Do not provide an answer—only reformulate the "
    "question if necessary; otherwise, return it unchanged."
))

This tells the model:

• It will receive a chat history and the latest user message.

• Its job is not to answer, but to reformulate the user’s question so it can stand alone.

• If the question already makes sense, return it unchanged.

✅ Example:

• Chat history:

  Human: Who is Elon Musk?
  AI: He is the CEO of Tesla.
  Human: What’s his age?
  

• Reformulated output:
  "What is the age of Elon Musk?"

4. MessagesPlaceholder("chat_history")

This acts as a dynamic placeholder for past messages (the conversation history).

When you later run this template (e.g. using .format() or .invoke()), you can insert actual chat messages here.

Example of runtime usage:

contextualize_question_prompt.format(
    chat_history=[
        HumanMessage(content="Who is Elon Musk?"),
        AIMessage(content="He is the CEO of Tesla.")
    ],
    input="What’s his age?"
)

→ The placeholder {chat_history} will be replaced by the two messages.

5. ("human", "{input}")

This defines the latest user message (the new query to contextualize).

{input} is a variable placeholder — it will be filled with the latest user input text during runtime.

For example:

input = "What’s his age?"

The prompt becomes:

Human: What’s his age?

🧩 Putting It All Together

At runtime, LangChain merges everything into a structured sequence like this:

System: Given a conversation history and the most recent user query, rewrite the query as a standalone question...
Human: [previous chat history messages...]
Human: What’s his age?

Then the LLM receives this complete prompt and outputs something like:

"What is the age of Elon Musk?"

🧾 Summary Table