Welcome to RAG Time Discussions! 🎉

[aycabas]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[sharmarohit5]

In Journey 1 RAG-Fundamentals notebook, step 6 is missng the code where USER_MESSAGE is augmented with search results retrieved in previous step. Shouldn't it be USER_MESSAGE = f"{retrieved_context}\n\n{user_question}"? or am I missing something?

[aycabas]

@farzad528 will you be able to help please?

[farzad528]

You are correct! The context was omitted being passed into the SYSTEM or USER_MESSAGE.

In RAG implementations, there are different ways to incorporate retrieved context. In this notebook, the retrieved context is being included in the SYSTEM_MESSAGE rather than the USER_MESSAGE. Personally, I prefer this approach for few reasons:

1. The system message is designed to provide context and instructions to the model
2. The user message should contain only what the user actually asked
3. Separating them maintains a cleaner conversation structure

[abbhakan]

Hello,

In the Journey 1 - RAG and Indexing Fundamentals there is a difference in the code shown in the video segment and the code in the notebook on Github. It seems that the step for augmenting the query with the search results has been omitted in the Github repo.

This is the code from the video:

First, let's collect the context from the search results

context = ""
for result in search_results:
context += result["chunk"] + "\n\n"

SYSTEM_MESSAGE = """
You are an AI assistant. Be brief in your answers.
Answer ONLY with the facts listed in the retreived text.

Context:
{context}
"""
USER_MESSAGE = user_question

Now we can use the matches to generate a response

response = openai_client.chat.completions.create(
model=os.getenv("AZURE_OPENAI_CHAT_COMPLETION_DEPLOYED_MODEL_NAME"),
temperature=0.7,
messages=[
{"role": "system", "content": SYSTEM_MESSAGE},
{"role": "user", "content": USER_MESSAGE},
],
)

This is the code from the Notebook on Github

SYSTEM_MESSAGE = """
Assistant helps company employees questions about the employee handbook. Be brief in your answers.
Answer ONLY with the facts listed in the retreived text.
"""
USER_MESSAGE = user_question

Now we can use the matches to generate a response

response = openai_client.chat.completions.create(
model=os.getenv("AZURE_OPENAI_CHAT_COMPLETION_DEPLOYED_MODEL_NAME"),
temperature=0.7,
messages=[
{"role": "system", "content": SYSTEM_MESSAGE},
{"role": "user", "content": USER_MESSAGE},
],
)

[farzad528]

The code in the sample is the latest most up to date code. Please refer this to that as the source of truth. Did slight refactoring to make the code cleaner.