Mini Project: Document Chatbot – Part 2 (Adding Memory)

Let's think this through for a moment

The RetrievalQA chain we built in Part 1 handles each question in isolation, so if you ask a follow-up like "can you explain that further?", there's no context to work with. In Part 2, we'll use the Memory concept from the Intermediate chapter to upgrade the chatbot into a conversational RAG chain. By using ConversationBufferMemory as chat_history and building a ConversationalRetrievalChain, we can keep the user's previous questions and answers stored in memory. The result is a chatbot the user can hold a back-and-forth conversation with about the document's content. Once this stage is done, we'll add agent tools in Part 3.

Let's build it

Reuse the vectorstore/retriever from Part 1. Create a ConversationBufferMemory(memory_key="chat_history", return_messages=True). Instead of RetrievalQA, build a new chain with ConversationalRetrievalChain.from_llm(llm, retriever, memory=memory). Ask "What is LangChain?" first, then follow up with "Can you give me an example of that?" and test whether it answers correctly by referencing memory. Print the result of chain.invoke() and observe how chat_history gets updated.

• Were you able to connect ConversationBufferMemory to the chain?
• Were you able to build a ConversationalRetrievalChain paired with the Part 1 retriever?
• Did you confirm that the follow-up question was answered correctly based on prior context?

Code Example

from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain

# Part 1 ကနေ vectorstore/retriever ကို ပြန်အသုံးပြု
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})

# Conversation memory တစ်ခု ဖန်တီး
memory = ConversationBufferMemory(
    memory_key="chat_history", return_messages=True
)

# Memory ပါတဲ့ conversational RAG chain ဆောက်
conv_chain = ConversationalRetrievalChain.from_llm(
    llm=llm,
    retriever=retriever,
    memory=memory,
)

# ပထမ မေးခွန်း
r1 = conv_chain.invoke({"question": "LangChain ဆိုတာဘာလဲ?"})
print("Answer 1:", r1["answer"])

# Follow-up မေးခွန်း (context ကို memory ကနေ ယူသုံးမည်)
r2 = conv_chain.invoke({"question": "အဲဒါနဲ့ ပတ်သက်ပြီး ဥပမာတစ်ခု ပေးပါ"})
print("Answer 2:", r2["answer"])

On the second question, the chatbot will correctly understand what 'that' refers to by drawing on the first question and answer stored in memory.