utils.py

from huggingface_hub import InferenceClient
from PIL import Image
import io
import base64
import json

def detect_food_or_ingredient(image: Image.Image, client: InferenceClient, api_key: str) -> dict:
    """Detect food or ingredient using a vision-capable LLM (e.g., LLaVA-13B)."""
    # Convert image to base64 for LLM processing
    buffered = io.BytesIO()
    image.save(buffered, format="JPEG")
    img_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")

    prompt = """
    You are an expert in Nigerian cuisine. Analyze the provided image and identify whether it depicts a prepared food dish or a raw ingredient. Return a JSON response with:
    - name: The name of the food or ingredient (e.g., 'jollof rice', 'yam')
    - type: Either 'food' or 'ingredient'
    Focus on Nigerian foods and ingredients (e.g., egusi soup, pounded yam, tomatoes, okra).
    """
    # Use chat completion for vision models like LLaVA
    response = client.chat_completion(
        messages=[
            {
                "role": "user",
                "content": [
                    {"type": "text", "text": prompt},
                    {"type": "image_url", "image_url": f"data:image/jpeg;base64,{img_base64}"}
                ]
            }
        ],
        max_tokens=100
    )
    result = json.loads(response.choices[0].message.content)
    return result

def get_nutritional_info(item: str, client: InferenceClient) -> str:
    """Get nutritional info and preparation methods using Nebius LLM."""
    prompt = f"""
    You are a nutrition expert specializing in Nigerian cuisine. Provide detailed information about {item}, including:
    - Health benefits
    - Preparation methods
    - Diabetic-friendly options
    - Ingredients breakdown
    Format the response in markdown with clear sections.
    """
    response = client.text_generation(prompt, max_new_tokens=500)
    return response

def suggest_recipe(ingredient: str, client: InferenceClient) -> dict:
    """Suggest a recipe based on an ingredient using Nebius LLM."""
    prompt = f"""
    You are a chef specializing in Nigerian cuisine. Given the ingredient {ingredient}, suggest a recipe that includes:
    - Dish name
    - Additional ingredients needed
    - Step-by-step preparation instructions
    Format the response in markdown with clear sections.
    """
    response = client.text_generation(prompt, max_new_tokens=500)
    return {"dish_name": f"{ingredient}-based dish", "recipe": response}

def generate_dish_image(dish_name: str, api_key: str) -> Image.Image:
    """Generate an image of the prepared dish using Nebius Flux.1."""
    client = InferenceClient(model="black-forest-labs/FLUX.1-schnell", token=api_key)
    image_bytes = client.text_to_image(f"A beautifully prepared Nigerian dish: {dish_name}")
    return Image.open(io.BytesIO(image_bytes))

def speech_to_text(audio_path: str) -> str:
    """Convert audio to text using Hugging Face Whisper model."""
    client = InferenceClient(model="openai/whisper-large-v3")
    with open(audio_path, "rb") as audio_file:
        text = client.audio_to_text(audio_file)
    return text