Working with Custom Data Sources

POST https://<your_openai_endpoint>/openai/deployments/<deployment>/chat/completions?api-version=<api_version>
Content-Type: application/json
api-key: <your_api_key>

{
  "data_sources": [
    {
      "type": "azure_search",
      "parameters": {
        "endpoint": "https://<your_search_endpoint>",
        "index_name": "<your_search_index>",
        "authentication": {
          "type": "system_assigned_managed_identity"
        },
        "semantic_configuration": "default",
        "query_type": "simple",
        "top_n_documents": 5,
        "strictness": 3,
        "role_information": "Answer only if you find relevant content in the data source. Do not guess. If unsure, say: \"I don't have information on that topic.\""
      }
    }
  ],
  "messages": [
    {
      "role": "system",
      "content": "Answer only if you find relevant content in the data source. Do not guess. If unsure, say: \"I don't have information on that topic.\""
    },
    {
      "role": "user",
      "content": "Who are the lead researchers for Project Orion?"
    }
  ],
  "past_messages": 10,
  "temperature": 0.0,
  "max_tokens": 800
}

# app.py
import os
from openai import AzureOpenAI
from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient

# Configuration (use environment variables in production)
AZURE_OPENAI_ENDPOINT = os.getenv("AZURE_OPENAI_ENDPOINT", "https://<your_openai_endpoint>")
AZURE_OPENAI_API_KEY = os.getenv("AZURE_OPENAI_API_KEY", "<your_openai_api_key>")
DEPLOYMENT_NAME = os.getenv("AZURE_OPENAI_DEPLOYMENT", "gpt-4o")

SEARCH_ENDPOINT = os.getenv("SEARCH_ENDPOINT", "https://<your_search_endpoint>")
SEARCH_KEY = os.getenv("SEARCH_KEY", "<your_search_key>")
SEARCH_INDEX_NAME = os.getenv("SEARCH_INDEX", "rag")

# Initialize Azure OpenAI client (key-based auth example)
client = AzureOpenAI(
    api_key=AZURE_OPENAI_API_KEY,
    azure_endpoint=AZURE_OPENAI_ENDPOINT,
    api_version="2024-02-15-preview"
)

def query_search_service(query_text, top_n=5):
    """Query Azure Cognitive Search and return top N documents."""
    search_client = SearchClient(
        endpoint=SEARCH_ENDPOINT,
        index_name=SEARCH_INDEX_NAME,
        credential=AzureKeyCredential(SEARCH_KEY)
    )

    results = search_client.search(query_text, top=top_n)
    docs = []
    for r in results:
        # r is a SearchResult; r.document contains the indexed fields
        docs.append(r.document)
    return docs

def format_documents_for_prompt(documents):
    """Convert search documents to a single context string for the model."""
    parts = []
    for i, doc in enumerate(documents, start=1):
        # Adjust fields according to your index schema. Example uses 'title' and 'content'
        title = doc.get("title", f"Document {i}")
        content = doc.get("content", "")
        parts.append(f"Source: {title}\n{content}\n")
    return "\n---\n".join(parts)

def ask_question_with_rag(question):
    # 1) Retrieve relevant documents
    docs = query_search_service(question, top_n=5)
    if not docs:
        return "I don't have information on that topic."

    # 2) Prepare context from documents
    context_block = format_documents_for_prompt(docs)

    # 3) Prepare messages (system instructs to only answer with supporting evidence)
    messages = [
        {
            "role": "system",
            "content": "Answer only if you find supporting content in the provided data. Do not guess. If unsure, say: \"I don't have information on that topic.\""
        },
        {
            "role": "user",
            "content": question
        },
        {
            "role": "system",
            "content": f"Context documents:\n{context_block}"
        }
    ]

    # 4) Send request to Azure OpenAI (RAG enabled via explicit context handling)
    response = client.chat.completions.create(
        model=DEPLOYMENT_NAME,
        messages=messages,
        max_tokens=800,
        temperature=0.0
    )

    # Extract the assistant message
    assistant_msg = response.choices[0].message["content"]
    return assistant_msg

def main():
    print("Azure OpenAI RAG Demo")
    print("----------------------")

    # Example question (replace with your prompt)
    question = "Who are the lead researchers for Project Orion?"
    print(f"\nQuestion: {question}")

    print("\nRetrieving information and generating answer...")
    answer = ask_question_with_rag(question)

    print("\nAnswer:")
    print(answer)

if __name__ == "__main__":
    main()

Azure OpenAI RAG Demo
----------------------

Question: Who are the lead researchers for Project Orion?

Retrieving information and generating answer...

Answer:
The lead researchers for Project Orion are:
- Dr. Eliza Tran (AI Systems Architect)
- Major Samuel Drake (Defense Operations Liaison)
- Ava Kohli (Azure Systems Engineer)