Handling follow-up questions in LangChain - Step-by-Step Execution

Concept Flow - Handling follow-up questions

User asks initial question

↓

LangChain processes question

↓

Generate answer

↓

User asks follow-up question

↓

LangChain uses context + follow-up

↓

Generate follow-up answer

↓

Repeat or end

This flow shows how LangChain handles an initial question, then uses context to answer follow-up questions.

Execution Sample

LangChain

from langchain.chains import ConversationChain
from langchain.llms import OpenAI

conversation = ConversationChain(llm=OpenAI())
print(conversation.predict(input="Hello!"))
print(conversation.predict(input="Can you tell me more?"))

This code creates a conversation chain that remembers context to answer follow-up questions.

Execution Table

Step	Input	Context Before	Action	Context After	Output
1	"Hello!"	{}	Process input, generate answer	{"history": ["Hello!"]}	"Hi! How can I help you today?"
2	"Can you tell me more?"	{"history": ["Hello!"]}	Use history + input to generate follow-up answer	{"history": ["Hello!", "Can you tell me more?"]}	"Sure! What would you like to know more about?"
3	"Thanks!"	{"history": ["Hello!", "Can you tell me more?"]}	Add input to context, generate closing answer	{"history": ["Hello!", "Can you tell me more?", "Thanks!"]}	"You're welcome! Let me know if you have more questions."
4	"Bye"	{"history": ["Hello!", "Can you tell me more?", "Thanks!"]}	Add input, generate farewell	{"history": ["Hello!", "Can you tell me more?", "Thanks!", "Bye"]}	"Goodbye! Have a great day!"
5	N/A	{"history": ["Hello!", "Can you tell me more?", "Thanks!", "Bye"]}	No more input, end conversation	{"history": ["Hello!", "Can you tell me more?", "Thanks!", "Bye"]}	Conversation ended

💡 Conversation ends when user stops sending input or says goodbye.

Variable Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	Final
history	[]	["Hello!"]	["Hello!", "Can you tell me more?"]	["Hello!", "Can you tell me more?", "Thanks!"]	["Hello!", "Can you tell me more?", "Thanks!", "Bye"]	["Hello!", "Can you tell me more?", "Thanks!", "Bye"]

Key Moments - 3 Insights

How does LangChain remember previous questions?

LangChain stores previous inputs in the 'history' context (see execution_table steps 1-4), so it can use them to understand follow-up questions.

What happens if the user asks a follow-up question without context?

LangChain uses the stored 'history' to provide context. Without history, it treats the input as a new question (see step 1).

Why does the context grow after each input?

Each user input is added to 'history' to keep track of the conversation flow, enabling better follow-up answers (see variable_tracker).

Visual Quiz - 3 Questions

Look at the execution_table at step 2. What is the context before processing the input?

A {}

B {"history": ["Can you tell me more?"]}

C {"history": ["Hello!"]}

D {"history": ["Hello!", "Can you tell me more?"]}

💡 Hint

Check the 'Context Before' column at step 2 in the execution_table.

At which step does the conversation end according to the execution_table?

A Step 4

B Step 5

C Step 3

D Step 2

💡 Hint

Look for the step where the output says 'Conversation ended'.

If the user sends a new question after 'Bye', how does the 'history' variable change?

A It adds the new question to the existing history

B It resets to empty []

C It removes 'Bye' from history

D It duplicates the last input

💡 Hint

Refer to the variable_tracker showing how 'history' grows after each input.

Concept Snapshot

Handling follow-up questions in LangChain:
- Use ConversationChain to keep context.
- Each input updates conversation history.
- Follow-up questions use previous context.
- Context stored in 'history' variable.
- Conversation ends when no input or user says goodbye.

Full Transcript

This visual execution shows how LangChain handles follow-up questions by keeping a conversation history. When a user asks an initial question, LangChain processes it and stores it in the history. For follow-up questions, LangChain uses this stored history to understand context and generate relevant answers. Each user input updates the history, allowing the conversation to flow naturally. The conversation ends when the user stops sending inputs or says goodbye. This approach helps LangChain remember what was said before and respond appropriately to follow-ups.

Practice

1. What is the main purpose of handling follow-up questions in LangChain conversations?

easy

A. To speed up the response time by skipping context

B. To keep the conversation natural and connected by remembering context

C. To reset the conversation after every question

D. To avoid using any previous conversation history

Solution

1. Step 1: Understand conversation flow

   Handling follow-up questions means the system remembers previous messages to keep the chat natural.

2. Step 2: Identify the purpose of context

   Context helps the system answer related questions without losing track of the topic.

3. Final Answer:

   To keep the conversation natural and connected by remembering context -> Option B

4. Quick Check:

   Follow-up questions need context = A [OK]

Hint: Follow-up means remembering past chat context [OK]

Common Mistakes:

• Thinking follow-ups reset conversation
• Ignoring conversation history
• Assuming follow-ups speed up responses
• Believing context is not needed

2. Which LangChain class is commonly used to manage conversation history for follow-up questions?

easy

A. ConversationChain

B. TextSplitter

C. DocumentLoader

D. VectorStoreRetriever

Solution

1. Step 1: Recall LangChain components

   ConversationChain is designed to handle conversations and keep track of history.

2. Step 2: Match class to purpose

   TextSplitter and DocumentLoader are for documents, VectorStoreRetriever for searching, not conversation management.

3. Final Answer:

   ConversationChain -> Option A

4. Quick Check:

   ConversationChain manages chat history = B [OK]

Hint: ConversationChain handles chat history in LangChain [OK]

Common Mistakes:

• Confusing ConversationChain with document tools
• Choosing unrelated classes like TextSplitter
• Assuming VectorStoreRetriever manages conversations
• Mixing up class purposes

3. Given this code snippet using LangChain's ConversationChain:

from langchain.chains import ConversationChain
from langchain.llms import OpenAI

llm = OpenAI(temperature=0)
conv = ConversationChain(llm=llm)

response1 = conv.predict(input="What is AI?")
response2 = conv.predict(input="Can you explain more?")
print(response2)

What does response2 represent?

medium

A. An error because ConversationChain does not support multiple predicts

B. An answer to 'Can you explain more?' ignoring previous questions

C. An answer to 'Can you explain more?' using the context of 'What is AI?'

D. The same answer as response1 repeated

Solution

1. Step 1: Understand ConversationChain behavior

   ConversationChain keeps track of previous inputs and outputs to provide context-aware answers.

2. Step 2: Analyze the code flow

   response2 uses the previous question 'What is AI?' to answer 'Can you explain more?' with context.

3. Final Answer:

   An answer to 'Can you explain more?' using the context of 'What is AI?' -> Option C

4. Quick Check:

   ConversationChain uses history for follow-ups = C [OK]

Hint: ConversationChain.predict uses past chat context automatically [OK]

Common Mistakes:

• Assuming each predict is independent
• Expecting an error on multiple calls
• Thinking response2 repeats response1
• Ignoring conversation history

4. Identify the error in this LangChain code snippet for handling follow-up questions:

from langchain.chains import ConversationChain
from langchain.llms import OpenAI

llm = OpenAI(temperature=0)
conv = ConversationChain(llm=llm)

response1 = conv.predict(input="Hello, who won the world cup?")
response2 = conv.predict("Who scored the most goals?")
print(response2)

medium

A. The print statement is invalid syntax

B. ConversationChain cannot be called twice

C. OpenAI LLM requires temperature > 0

D. Missing 'input=' keyword in second predict call causes error

Solution

1. Step 1: Check method call syntax

   ConversationChain.predict expects a named argument 'input=...'. The second call misses this keyword.

2. Step 2: Understand consequences

   Without 'input=', Python raises a TypeError because the argument is positional but not expected.

3. Final Answer:

   Missing 'input=' keyword in second predict call causes error -> Option D

4. Quick Check:

   predict requires input= keyword = D [OK]

Hint: Always use input= keyword in predict calls [OK]

Common Mistakes:

• Calling predict without input= keyword
• Thinking ConversationChain disallows multiple calls
• Believing temperature=0 is invalid
• Assuming print syntax is wrong

5. You want to build a LangChain app that answers follow-up questions about a user's previous queries but resets context after 3 questions. Which approach best achieves this?

hard

A. Use ConversationChain and clear its memory after every 3 predictions

B. Create a new ConversationChain instance for each question

C. Use ConversationChain without clearing memory to keep all history

D. Use OpenAI directly without ConversationChain

Solution

1. Step 1: Understand requirement to reset context

   The app needs to keep context but reset after 3 questions to avoid long history.

2. Step 2: Choose method to reset memory

   ConversationChain supports memory clearing; clearing after 3 predictions resets context properly.

3. Step 3: Evaluate other options

   Creating new instances loses all context immediately; keeping all history ignores reset; using OpenAI directly lacks conversation memory.

4. Final Answer:

   Use ConversationChain and clear its memory after every 3 predictions -> Option A

5. Quick Check:

   Reset context by clearing ConversationChain memory = A [OK]

Hint: Clear ConversationChain memory to reset context after set questions [OK]

Common Mistakes:

• Recreating ConversationChain loses partial context
• Not resetting memory causes long context buildup
• Using OpenAI directly loses conversation flow
• Ignoring how to clear memory properly