🏋️ Fun with Text and Image Embeddings 🍎¶

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 ImageEmbeddingInput

# Load environment variables (optional, if you keep them in a .env file)
load_dotenv()

# Retrieve from environment or fallback
PROJECT_CONNECTION_STRING = os.environ.get("PROJECT_CONNECTION_STRING", "<your-connection-string>")
MODEL_DEPLOYMENT_NAME = os.environ.get("MODEL_DEPLOYMENT_NAME", "<your-deployment-name>")

# Initialize project client
try:
    project_client = AIProjectClient.from_connection_string(
        credential=DefaultAzureCredential(),
        conn_str=PROJECT_CONNECTION_STRING,
    )
    print("🎉 Successfully created AIProjectClient")
except Exception as e:
    print("❌ Error creating AIProjectClient:", e)

text_phrases = [
    "An apple a day keeps the doctor away 🍎",
    "Quick 15-minute HIIT workout routine 🏋️",
    "Mindful breathing exercises 🧘"
]

try:
    with project_client.inference.get_embeddings_client() as embed_client:
        response = embed_client.embed(
            model=MODEL_DEPLOYMENT_NAME,
            input=text_phrases
        )

        for item in response.data:
            vec = item.embedding
            sample_str = f"[{vec[0]:.4f}, {vec[1]:.4f}, ..., {vec[-2]:.4f}, {vec[-1]:.4f}]"
            print(f"Sentence {item.index}: '{text_phrases[item.index]}':\n" \
                  f"  Embedding length={len(vec)}\n" \
                  f"  Sample: {sample_str}\n")
except Exception as e:
    print("❌ Error embedding text:", e)

# A basic prompt template (system-style) we'll prepend to user text.
TEMPLATE_SYSTEM = (
    "You are HealthFitGPT, a fitness guidance model.\n"
    "Please focus on healthy advice and disclaim you're not a doctor.\n\n"
    "User message:"  # We'll append user message after this.
)

def embed_with_template(user_text):
    """Embed user text with a system template in front."""
    content = TEMPLATE_SYSTEM + user_text
    with project_client.inference.get_embeddings_client() as embed_client:
        rsp = embed_client.embed(
            model=MODEL_DEPLOYMENT_NAME,
            input=[content]
        )
    return rsp.data[0].embedding

sample_user_text = "Can you suggest a quick home workout for busy moms?"
embedding_result = embed_with_template(sample_user_text)
print("Embedding length:", len(embedding_result))
print("First few dims:", embedding_result[:8])

image_file_path = "gbb.jpeg"  # Replace with your local image path!
try:
    with project_client.inference.get_image_embeddings_client() as img_embed_client:
        # Construct input for the image embeddings call
        img_input = ImageEmbeddingInput.load(image_file=image_file_path, image_format="jpg")
        resp = img_embed_client.embed(
            model=MODEL_DEPLOYMENT_NAME,
            input=[img_input]
        )

        for item in resp.data:
            vec = item.embedding
            snippet = f"[{vec[0]:.4f}, {vec[1]:.4f}, ..., {vec[-2]:.4f}, {vec[-1]:.4f}]"
            print(f"Image index={item.index}, length={len(vec)}, sample={snippet}")
except Exception as e:
    print("❌ Error embedding image:", e)

import base64
from IPython.display import Image, display

# This block is pseudo-code / example, depending on your actual method names:
def generate_health_image(prompt="A simple cartoon of a happy person jogging outdoors"):
    try:
        with project_client.inference.get_image_generation_client() as gen_client:
            gen_response = gen_client.generate(
                model=MODEL_DEPLOYMENT_NAME,
                prompt=prompt,
                # possibly other params like size, etc.
            )
            # We'll assume gen_response has a base64 image or raw bytes in .data[0].
            # This is just an example structure:
            if gen_response.data:
                image_bytes = gen_response.data[0].binary  # or .content, .image_bytes, etc.
                return image_bytes
    except Exception as e:
        print("❌ Error generating image:", e)
    return None

# Let's do a quick run
generated_img_data = generate_health_image()
if generated_img_data:
    display(Image(generated_img_data))
else:
    print("(No generated image data. This may be stub code for demonstration.)")