🍎 Chat Completions with AIProjectClient 🍏¶

In this notebook, we'll demonstrate how to perform Chat Completions using the Azure AI Foundry SDK. We'll combine azure-ai-projects and azure-ai-inference packages to:

Initialize an AIProjectClient.
Obtain a Chat Completions client to do direct LLM calls.
Use a prompt template to add system context.
Send user prompts in a health & fitness theme.

🏋️ Health-Fitness Disclaimer¶

This example is for demonstration only and does not provide real medical advice. Always consult a professional for health or medical-related questions.

Prerequisites¶

Python 3.8+
azure-ai-projects, azure-ai-inference, azure-identity

A .env file containing:

PROJECT_CONNECTION_STRING=<your-project-connection-string>
MODEL_DEPLOYMENT_NAME=<your-model-deployment-name>

Let's get started! 🎉

1. Initial Setup¶

Load environment variables, create an AIProjectClient, and fetch a ChatCompletionsClient. We'll also define a prompt template to show how you might structure a system message.

In [ ]:

  Copied!     
 
import os
from dotenv import load_dotenv
from pathlib import Path
from azure.identity import DefaultAzureCredential

from azure.ai.projects import AIProjectClient
from azure.ai.inference.models import UserMessage, SystemMessage  # for chat messages

# Load environment variables
notebook_path = Path().absolute()
parent_dir = notebook_path.parent
load_dotenv(parent_dir / '.env')

# Retrieve from environment
connection_string = os.environ.get("PROJECT_CONNECTION_STRING")
model_deployment = os.environ.get("MODEL_DEPLOYMENT_NAME")

try:
    # Create the project client
    project_client = AIProjectClient.from_connection_string(
        credential=DefaultAzureCredential(),
        conn_str=connection_string,
    )
    print("✅ Successfully created AIProjectClient")
except Exception as e:
    print("❌ Error initializing client:", e)
import os from dotenv import load_dotenv from pathlib import Path from azure.identity import DefaultAzureCredential from azure.ai.projects import AIProjectClient from azure.ai.inference.models import UserMessage, SystemMessage # for chat messages # Load environment variables notebook_path = Path().absolute() parent_dir = notebook_path.parent load_dotenv(parent_dir / '.env') # Retrieve from environment connection_string = os.environ.get("PROJECT_CONNECTION_STRING") model_deployment = os.environ.get("MODEL_DEPLOYMENT_NAME") try: # Create the project client project_client = AIProjectClient.from_connection_string( credential=DefaultAzureCredential(), conn_str=connection_string, ) print("✅ Successfully created AIProjectClient") except Exception as e: print("❌ Error initializing client:", e) 

Prompt Template¶

We'll define a quick system message that sets the context as a friendly, disclaimer-providing fitness assistant.

txt
SYSTEM PROMPT (template):
You are FitChat GPT, a helpful fitness assistant.
Always remind users: I'm not a medical professional.
Be friendly, provide general advice.
...

We'll then pass user content as a user message.

In [ ]:

  Copied!     
 
# We'll define a function that runs chat completions with a system prompt & user prompt
def chat_with_fitness_assistant(user_input: str):
    """Use chat completions to get a response from our LLM, with system instructions."""
    # Our system message template
    system_text = (
        "You are FitChat GPT, a friendly fitness assistant.\n"
        "Always remind users: I'm not a medical professional.\n"
        "Answer with empathy and disclaimers."
    )

    # We'll open the chat completions client
    with project_client.inference.get_chat_completions_client() as chat_client:
        # Construct messages: system + user
        system_message = SystemMessage(content=system_text)
        user_message = UserMessage(content=user_input)

        # Send the request
        response = chat_client.complete(
            model=model_deployment,
            messages=[system_message, user_message]
        )

        return response.choices[0].message.content  # simplest approach: get top choice's content

print("Defined a helper function to do chat completions.")
# We'll define a function that runs chat completions with a system prompt & user prompt def chat_with_fitness_assistant(user_input: str): """Use chat completions to get a response from our LLM, with system instructions.""" # Our system message template system_text = ( "You are FitChat GPT, a friendly fitness assistant.\n" "Always remind users: I'm not a medical professional.\n" "Answer with empathy and disclaimers." ) # We'll open the chat completions client with project_client.inference.get_chat_completions_client() as chat_client: # Construct messages: system + user system_message = SystemMessage(content=system_text) user_message = UserMessage(content=user_input) # Send the request response = chat_client.complete( model=model_deployment, messages=[system_message, user_message] ) return response.choices[0].message.content # simplest approach: get top choice's content print("Defined a helper function to do chat completions.")

2. Try Chat Completions 🎉¶

We'll call the function with a user question about health or fitness, and see the result. Feel free to modify the question or run multiple times!

In [ ]:

  Copied!     
 
user_question = "How can I start a beginner workout routine at home?"
reply = chat_with_fitness_assistant(user_question)
print("🗣️ User:", user_question)
print("🤖 Assistant:", reply)
user_question = "How can I start a beginner workout routine at home?" reply = chat_with_fitness_assistant(user_question) print("🗣️ User:", user_question) print("🤖 Assistant:", reply)

3. Another Example: Prompt Template with Fill-Ins 📝¶

We can go a bit further and add placeholders in the system message. For instance, imagine we have a userName or goal. We'll show a minimal example.

In [ ]:

  Copied!     
 
def chat_with_template(user_input: str, user_name: str, goal: str):
    # Construct a system template with placeholders
    system_template = (
        "You are FitChat GPT, an AI personal trainer for {name}.\n"
        "Your user wants to achieve: {goal}.\n"
        "Remind them you're not a medical professional. Offer friendly advice."
    )

    # Fill in placeholders
    system_prompt = system_template.format(name=user_name, goal=goal)

    with project_client.inference.get_chat_completions_client() as chat_client:
        system_msg = SystemMessage(content=system_prompt)
        user_msg = UserMessage(content=user_input)

        response = chat_client.complete(
            model=model_deployment,
            messages=[system_msg, user_msg]
        )

    return response.choices[0].message.content

# Let's try it out
templated_user_input = "What kind of home exercise do you recommend for a busy schedule?"
assistant_reply = chat_with_template(
    templated_user_input,
    user_name="Jordan",
    goal="increase muscle tone and endurance"
)
print("🗣️ User:", templated_user_input)
print("🤖 Assistant:", assistant_reply)
def chat_with_template(user_input: str, user_name: str, goal: str): # Construct a system template with placeholders system_template = ( "You are FitChat GPT, an AI personal trainer for {name}.\n" "Your user wants to achieve: {goal}.\n" "Remind them you're not a medical professional. Offer friendly advice." ) # Fill in placeholders system_prompt = system_template.format(name=user_name, goal=goal) with project_client.inference.get_chat_completions_client() as chat_client: system_msg = SystemMessage(content=system_prompt) user_msg = UserMessage(content=user_input) response = chat_client.complete( model=model_deployment, messages=[system_msg, user_msg] ) return response.choices[0].message.content # Let's try it out templated_user_input = "What kind of home exercise do you recommend for a busy schedule?" assistant_reply = chat_with_template( templated_user_input, user_name="Jordan", goal="increase muscle tone and endurance" ) print("🗣️ User:", templated_user_input) print("🤖 Assistant:", assistant_reply)

4. Tips & Cleanup¶

Tips:

Add OpenTelemetry for end-to-end tracing of your calls.
Evaluate your chat completions with azure-ai-evaluation to see how well your LLM is performing.
Combine system + user + possibly assistant messages if you want to pass prior conversation context.

No Cleanup Required¶

Because we're using a direct chat completions client, we haven't created any persistent resources we need to tear down. In more advanced scenarios, you'd manage resources like Agents, Threads, or Tools.

🎉 Congratulations!¶

You've successfully performed chat completions with the Azure AI Foundry's AIProjectClient and azure-ai-inference. You've also seen how to incorporate prompt templates to tailor your system instructions.

Keep exploring, and stay healthy & fit! 🏋️