Introduction

In April 2025, Microsoft announced the integration of Azure OpenAI Service into Azure AI Foundry — and as someone who’s been building with AWS Bedrock, I immediately noticed how similar it looked. The new Azure AI Foundry platform offers an interface where you can test prompts, add instructions, and interact with your deployed models — no extra infra needed.

So, I thought — why not build a full-stack AI app that combines Apple Health data with GPT-4o on Azure?

This became AI Health Agent, an iOS app where you can ask questions like:

“How active was I last month?”

“Is my recent weight trend okay?”

“How can I gradually improve my heart health?”

And get intelligent answers grounded in your actual health metrics—no backend, no database. Just you, your device, and the power of GPT. I use ChatGPT as my daily weight loss coach, so I know it can do a great job. Since I already store my health data in the Health app, I realized how powerful it would be if ChatGPT could access and analyze that data directly.

The app is on TestFlight for Demo use.

https://testflight.apple.com/join/xBj899wE

What It Does

The app reads your Apple Health data — like weight, step count, heart rate, and energy burned — and sends it (securely, and temporarily) to the Azure OpenAI API. The model replies as a friendly, knowledgeable assistant — like a cross between a personal trainer and a general physician.

It’s great for people who want to casually reflect on their wellness, without needing a clinical dashboard or medical jargon. Azure and Open AI integration is pritty simple, you need to create Azure AI service group and it leads to the Azure AI portal.

Prompt Engineering

Azure AI Foundry lets you define system instructions directly in the UI — just like AWS Bedrock. You can simulate full chat conversations, add temperature, and test edge cases right from the browser. But in my app, I define the full prompt structure in code.

Here’s a sample Python-style request using Azure SDK:

response = client.chat.completions.create(
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "I am going to Paris, what should I see?"},
        {"role": "assistant", "content": "Paris is known for..."},
        {"role": "user", "content": "What is so great about #1?"}
    ],
    max_tokens=4096,
    temperature=1.0,
    top_p=1.0,
    model=deployment
)

Here’s how I adapted it in Swift: