Why conversation history improves RAG in LangChain - Visual Breakdown

Concept Flow - Why conversation history improves RAG

User asks question

↓

Retrieve conversation history

↓

Combine history + new question

↓

Use combined input to query knowledge base

↓

Retrieve relevant documents

↓

Generate answer using retrieved docs + context

↓

Return improved, context-aware answer

The system uses past conversation to add context, then retrieves documents and generates a better answer.

Execution Sample

LangChain

history = ['What is AI?', 'Explain RAG.']
new_question = 'How does history help?'
combined_input = history + [new_question]
docs = retriever(combined_input)
answer = generator(docs, combined_input)
print(answer)

This code shows how conversation history is combined with a new question to retrieve documents and generate an answer.

Execution Table

Step	Action	Input	Output	Notes
1	Receive new question	'How does history help?'	new_question = 'How does history help?'	User asks a new question
2	Retrieve conversation history	['What is AI?', 'Explain RAG.']	history = ['What is AI?', 'Explain RAG.']	Past questions stored
3	Combine history and new question	history + [new_question]	combined_input = ['What is AI?', 'Explain RAG.', 'How does history help?']	Context is built
4	Retrieve documents using combined input	combined_input	docs = ['doc1', 'doc2']	Relevant docs found based on full context
5	Generate answer using docs and combined input	docs + combined_input	answer = 'History helps by providing context...'	Answer is context-aware
6	Output answer	answer	'History helps by providing context...'	Final improved answer shown

💡 All steps complete, answer generated with conversation history improving relevance.

Variable Tracker

Variable	Start	After Step 2	After Step 3	After Step 4	After Step 5	Final
history	[]	['What is AI?', 'Explain RAG.']	['What is AI?', 'Explain RAG.']	['What is AI?', 'Explain RAG.']	['What is AI?', 'Explain RAG.']	['What is AI?', 'Explain RAG.']
new_question	''	''	'How does history help?'	'How does history help?'	'How does history help?'	'How does history help?'
combined_input	[]	[]	['What is AI?', 'Explain RAG.', 'How does history help?']	['What is AI?', 'Explain RAG.', 'How does history help?']	['What is AI?', 'Explain RAG.', 'How does history help?']	['What is AI?', 'Explain RAG.', 'How does history help?']
docs	[]	[]	[]	['doc1', 'doc2']	['doc1', 'doc2']	['doc1', 'doc2']
answer	''	''	''	''	'History helps by providing context...'	'History helps by providing context...'

Key Moments - 3 Insights

Why do we combine conversation history with the new question before retrieving documents?

Combining history with the new question (see Step 3 in execution_table) gives the retriever more context, so it finds documents that better match the full conversation, not just the last question.

What happens if we don't use conversation history in retrieval?

If history is not used, the retriever only sees the new question (Step 1), so it might miss important context, leading to less relevant documents and a weaker answer (compare Step 4 with and without history).

How does the answer generation use the conversation history?

The answer generator uses both the retrieved documents and the combined input (Step 5) to produce an answer that fits the whole conversation, making it more accurate and coherent.

Visual Quiz - 3 Questions

Look at the execution_table at Step 3. What does combined_input contain?

A Only the new question

B Only the conversation history

C Conversation history plus the new question

D Retrieved documents

💡 Hint

Check the 'Output' column at Step 3 in the execution_table.

At which step does the system retrieve documents based on the combined input?

A Step 2

B Step 4

C Step 5

D Step 6

💡 Hint

Look for the step where 'Retrieve documents' is the action in the execution_table.

If we removed conversation history, how would the 'docs' variable change at Step 4?

A It would contain more documents

B It would be empty

C It might contain less relevant documents

D It would be the same

💡 Hint

Refer to the key_moments explanation about the importance of history in retrieval.

Concept Snapshot

Why conversation history improves RAG:
- Combine past conversation with new question
- Use combined input to retrieve documents
- Generate answer using docs + full context
- History adds important context
- Leads to more relevant retrieval and better answers

Full Transcript

This visual execution shows how conversation history improves Retrieval-Augmented Generation (RAG). First, the system receives a new question. Then it retrieves past conversation history. It combines the history with the new question to create a richer input. This combined input is used to retrieve relevant documents from the knowledge base. Finally, the answer generator uses both the retrieved documents and the combined input to produce a context-aware answer. Using conversation history helps the system understand the question better and find more relevant information, resulting in improved answers.

Practice

1. Why does including conversation history improve Retrieval-Augmented Generation (RAG) systems?

easy

A. It provides context that helps retrieve more relevant documents.

B. It reduces the size of the knowledge base.

C. It speeds up the model training process.

D. It removes the need for document retrieval.

Solution

1. Step 1: Understand the role of conversation history in RAG

   Conversation history gives the system context about what was previously discussed.

2. Step 2: Connect context to document retrieval

   With context, the system can find documents that better match the user's current question.

3. Final Answer:

   It provides context that helps retrieve more relevant documents. -> Option A

4. Quick Check:

   Context improves retrieval = A [OK]

Hint: Remember: context helps find better info [OK]

Common Mistakes:

• Thinking history reduces knowledge base size
• Confusing training speed with retrieval quality
• Assuming history removes retrieval step

2. Which of the following is the correct way to represent conversation history in Langchain for RAG?

easy

A. An integer count of messages exchanged.

B. A single string with all messages concatenated without structure.

C. A dictionary with only the last user message.

D. A structured list of message objects, each with role and content.

Solution

1. Step 1: Identify the recommended data structure for conversation history

   Langchain uses a list of message objects, each containing role (user/assistant) and content.

2. Step 2: Compare options to this structure

   Only structured list of message objects, each with role and content. matches this structured list format.

3. Final Answer:

   A structured list of message objects, each with role and content. -> Option D

4. Quick Check:

   Structured list of messages = A [OK]

Hint: Look for structured list with roles and content [OK]

Common Mistakes:

• Using unstructured strings
• Keeping only last message
• Using counts instead of messages

3. Given this simplified Langchain code snippet for RAG:

history = [
  {"role": "user", "content": "What is AI?"},
  {"role": "assistant", "content": "AI means Artificial Intelligence."}
]
query = "Explain AI applications."
retrieved_docs = retriever.get_relevant_documents(history + [{"role": "user", "content": query}])
print(len(retrieved_docs))

If the retriever returns 3 documents when given full history plus query, but only 1 document when given query alone, what does this show?

medium

A. The retriever ignores conversation history.

B. Conversation history helps retrieve more relevant documents.

C. Conversation history reduces the number of documents retrieved.

D. The retriever always returns the same number of documents.

Solution

1. Step 1: Compare retrieval results with and without history

   With history + query, 3 documents are retrieved; with query alone, only 1 document is retrieved.

2. Step 2: Interpret the effect of conversation history

   More documents retrieved with history means it helps find more relevant info.

3. Final Answer:

   Conversation history helps retrieve more relevant documents. -> Option B

4. Quick Check:

   More docs with history = C [OK]

Hint: More docs with history means better retrieval [OK]

Common Mistakes:

• Assuming fewer docs means better retrieval
• Ignoring difference in document counts
• Thinking retriever ignores history

4. You have this Langchain snippet for conversation history:

history = [
  {"role": "user", "content": "Hello"},
  {"role": "assistant", "content": "Hi! How can I help?"}
]
# Later code
retriever.get_relevant_documents(history)

Why might this cause an error or poor retrieval?

medium

A. Because the retriever expects a query string, not a list of messages.

B. Because the history list is empty.

C. Because the roles are misspelled.

D. Because the retriever only accepts integers.

Solution

1. Step 1: Check expected input type for retriever

   Retriever expects a query string or text, not a list of message objects.

2. Step 2: Identify mismatch in input

   Passing the entire history list causes errors or irrelevant retrieval.

3. Final Answer:

   Because the retriever expects a query string, not a list of messages. -> Option A

4. Quick Check:

   Retriever needs string query, not list [OK]

Hint: Retriever input must be a string query [OK]

Common Mistakes:

• Passing list instead of string
• Assuming empty history causes error
• Ignoring input type requirements

5. In a RAG system using Langchain, you want to improve answer quality by using conversation history. Which approach best achieves this?

hard

A. Concatenate all previous messages into one string and pass it as the query to the retriever.

B. Ignore conversation history and only use the current query for retrieval.

C. Use a structured list of past messages and append the current user query before retrieval.

D. Clear conversation history after every retrieval to avoid confusion.

Solution

1. Step 1: Understand how to use conversation history effectively

   Using a structured list preserves roles and context, helping retrieval.

2. Step 2: Combine history with current query properly

   Appending the current query to the history list keeps context and focuses retrieval.

3. Final Answer:

   Use a structured list of past messages and append the current user query before retrieval. -> Option C

4. Quick Check:

   Structured history + query = B [OK]

Hint: Keep structured history and add current query [OK]

Common Mistakes:

• Passing all messages as one string
• Ignoring history completely
• Clearing history too soon