app.py

import gradio as gr
from transformers import pipeline
from PIL import Image
import torch
import os
import spaces

# --- Initialize the Model Pipeline (No changes) ---
print("Loading MedGemma model...")
try:
    pipe = pipeline(
        "image-to-text",
        model="google/medgemma-4b-it",
        torch_dtype=torch.bfloat16,
        device_map="auto",
        token=os.environ.get("HF_TOKEN")
    )
    model_loaded = True
    print("Model loaded successfully!")
except Exception as e:
    model_loaded = False
    print(f"Error loading model: {e}")

# --- Core Analysis Function (Final Corrected Version) ---
@spaces.GPU()
def analyze_symptoms(symptom_image: Image.Image, symptoms_text: str):
    """
    Analyzes user's symptoms using the definitive correct calling convention for
    multimodal chat models in Hugging Face pipelines.
    """
    if not model_loaded:
        return "Error: The AI model could not be loaded. Please check the Space logs."

    symptoms_text = symptoms_text.strip() if symptoms_text else ""
    if symptom_image is None and not symptoms_text:
        return "Please describe your symptoms or upload an image for analysis."

    try:
        # --- DEFINITIVE CHAT-BASED PROMPT LOGIC ---
        system_instruction = (
            "You are an expert, empathetic AI medical assistant. "
            "Analyze the potential medical condition based on the user's input. "
            "Provide a list of possible conditions, your reasoning, and a clear, "
            "actionable next-steps plan. Begin your analysis by describing the information "
            "the user provided."
        )

        # The user's content is a list of dictionaries describing the multimodal input.
        user_content = []
        
        # The text part of the prompt.
        user_content.append({"type": "text", "text": symptoms_text})

        # The images to be passed to the pipeline.
        images_to_pass = []
        if symptom_image:
            # 1. Add a placeholder to the content list.
            user_content.append({"type": "image"})
            # 2. Add the actual PIL image object to a separate list.
            images_to_pass.append(symptom_image)

        # Construct the final messages list for the chat template.
        messages = [
            {"role": "system", "content": system_instruction},
            {"role": "user", "content": user_content},
        ]
        
        print("Generating pipeline output with separate image argument...")

        # Generation parameters must be in a `generate_kwargs` dictionary.
        generate_kwargs = {
            "max_new_tokens": 512,
            "do_sample": True,
            "temperature": 0.7,
        }
        
        # --- THE DEFINITIVE PIPELINE CALL ---
        # The pipeline call requires three distinct things:
        # 1. The `messages` list describing the chat structure.
        # 2. The `images` list containing the actual PIL image data.
        # 3. The `generate_kwargs` dictionary for generation parameters.
        output = pipe(
            messages,
            images=images_to_pass,
            generate_kwargs=generate_kwargs
        )

        print("Pipeline Output:", output)

        # --- SIMPLIFIED OUTPUT PROCESSING ---
        # For this pipeline, the output is a list with one dictionary.
        # The 'generated_text' key contains the AI's response string directly.
        if output and isinstance(output, list) and 'generated_text' in output[0]:
            result = output[0]['generated_text']
        else:
            result = "The model did not return a valid response. Please try again."

        disclaimer = "\n\n***Disclaimer: I am an AI assistant and not a medical professional. This is not a diagnosis. Please consult a doctor for any health concerns.***"
        
        return result + disclaimer

    except Exception as e:
        print(f"An exception occurred during analysis: {type(e).__name__}: {e}")
        return f"An error occurred during analysis. Please check the logs for details: {str(e)}"

# --- Gradio Interface (No changes needed) ---
with gr.Blocks(theme=gr.themes.Soft(), title="AI Symptom Analyzer") as demo:
    gr.HTML("""
        <div style="text-align: center; background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); color: white; padding: 2rem; border-radius: 10px; margin-bottom: 2rem;">
            <h1>🩺 AI Symptom Analyzer</h1>
            <p>Advanced symptom analysis powered by Google's MedGemma AI</p>
        </div>
    """)
    gr.HTML("""
        <div style="background-color: #fff3cd; border: 1px solid #ffeaa7; border-radius: 8px; padding: 1rem; margin: 1rem 0; color: #856404;">
            <strong>⚠️ Medical Disclaimer:</strong> This AI tool is for informational purposes only and is not a substitute for professional medical diagnosis or treatment.
        </div>
    """)
    
    with gr.Row(equal_height=True):
        with gr.Column(scale=1):
            gr.Markdown("### 1. Describe Your Symptoms")
            symptoms_input = gr.Textbox(
                label="Symptoms",
                placeholder="e.g., 'I have a rash on my arm that is red and itchy...'", lines=5)
            gr.Markdown("### 2. Upload an Image (Optional)")
            image_input = gr.Image(label="Symptom Image", type="pil", height=300)
            with gr.Row():
                clear_btn = gr.Button("🗑️ Clear All", variant="secondary")
                analyze_btn = gr.Button("🔍 Analyze Symptoms", variant="primary", size="lg")
                
        with gr.Column(scale=1):
            gr.Markdown("### 📊 Analysis Report")
            output_text = gr.Textbox(
                label="AI Analysis", lines=25, show_copy_button=True, placeholder="Analysis results will appear here...")

    def clear_all():
        return None, "", ""

    analyze_btn.click(fn=analyze_symptoms, inputs=[image_input, symptoms_input], outputs=output_text)
    clear_btn.click(fn=clear_all, outputs=[image_input, symptoms_input, output_text])

if __name__ == "__main__":
    print("Starting Gradio interface...")
    demo.launch(debug=True)