app.py

import gradio as gr
from google import genai
import os
import re
from PIL import Image, ImageDraw
import requests
from io import BytesIO

# --- 1. Custom CSS ---
custom_css = """
.improving { background-color: #dcf2dc; color: #155724; padding: 12px; border-radius: 8px; font-weight: bold; font-size: 1.2em; text-align: center; border: 1px solid #c3e6cb;}
.declining { background-color: #f8d7da; color: #721c24; padding: 12px; border-radius: 8px; font-weight: bold; font-size: 1.2em; text-align: center; border: 1px solid #f5c6cb;}
.stable { background-color: #fff3cd; color: #856404; padding: 12px; border-radius: 8px; font-weight: bold; font-size: 1.2em; text-align: center; border: 1px solid #ffeeba;}
"""

# --- 2. Google Maps API Fetcher ---
def fetch_streetview(address):
    """Geocodes an address and fetches the most recent Street View image."""
    gmaps_api_key = os.environ.get("GMAPS_API_KEY")
    if not gmaps_api_key:
        raise gr.Error("ERROR: GMAPS_API_KEY not found in Secrets.")
    if not address:
        raise gr.Error("Please enter a valid address.")
        
    gr.Info("Geocoding address...")
    
    # 1. Geocoding API
    geo_url = "https://maps.googleapis.com/maps/api/geocode/json"
    geo_params = {"address": address, "key": gmaps_api_key}
    geo_data = requests.get(geo_url, params=geo_params).json()
    
    if geo_data['status'] != 'OK':
        error_detail = geo_data.get('error_message', geo_data['status'])
        raise gr.Error(f"Maps API Error: {error_detail}")
        
    lat = geo_data['results'][0]['geometry']['location']['lat']
    lng = geo_data['results'][0]['geometry']['location']['lng']
    
    gr.Info("Fetching Street View imagery...")
    
    # 2. Street View Static API
    sv_url = "https://maps.googleapis.com/maps/api/streetview"
    sv_params = {
        "size": "640x640",
        "location": f"{lat},{lng}",
        "fov": "80",
        "pitch": "0",
        "key": gmaps_api_key
    }
    sv_res = requests.get(sv_url, params=sv_params)
    
    # --- ALIGNED IF/ELSE BLOCK ---
    if sv_res.status_code == 200:
        gr.Info("Image successfully loaded!")
        return Image.open(BytesIO(sv_res.content))
    else:
        error_msg = sv_res.text if sv_res.text else "No reason provided."
        raise gr.Error(f"Maps 403 Error: {error_msg}")
# --- 3. Spatial Drawing Engine ---
def draw_forensic_targets(image, report_text):
    annotated_img = image.copy()
    draw = ImageDraw.Draw(annotated_img)
    width, height = annotated_img.size
    
    coords = re.findall(r'\[(\d+),\s*(\d+)\]', report_text)
    for y_str, x_str in coords:
        y = int((int(y_str) / 1000) * height)
        x = int((int(x_str) / 1000) * width)
        r = 15 
        draw.ellipse((x - r, y - r, x + r, y + r), outline="red", width=5)
        
    return annotated_img

# --- 4. Robotics-ER Core Engine ---
def ground_truth_engine(past_img, present_img, audit_level, progress=gr.Progress()):
    progress(0, desc="Waking Spatial Sentinel...")
    
    api_key = os.environ.get("GOOGLE_API_KEY")
    if not api_key:
        return None, "<div class='declining'>ERROR: API Key Missing</div>", "Set GOOGLE_API_KEY in Secrets."

    client = genai.Client(api_key=api_key)
    if past_img is None or present_img is None:
        return None, "<div class='stable'>Missing Images</div>", "Please upload both images."

    audit_directives = {
        "Standard Audit": "Perform a structural comparison of roofing, siding, and landscaping.",
        "Deep Forensic": """Perform an exhaustive spatial audit. 
        Identify subtle wear (shingles, foundation, paint) and biological encroachment. 
        MANDATORY: Return center points for top 3 changes as [y, x] normalized coordinates (0-1000)."""
    }
    
    directive = audit_directives.get(audit_level)
    progress(0.2, desc=f"Running {audit_level}...")

    prompt = f"""
    SYSTEM INSTRUCTION: You are a structural forensic agent.
    DIRECTIVE: {directive}
    
    TASK: Compare 'Past Condition' vs 'Current Condition'. 
    Return a structured report. YOU MUST CONCLUDE with exactly one of these phrases: 
    'Maintenance Trajectory: IMPROVING', 'Maintenance Trajectory: STABLE', or 'Maintenance Trajectory: DECLINING'.
    """

    try:
        progress(0.5, desc="Robotics-ER is scanning for spatial anomalies...")
        response = client.models.generate_content(
            model="gemini-robotics-er-1.5-preview",
            contents=[prompt, past_img, present_img]
        )
        text_out = response.text
        
        badge_html = "<div class='stable'>➖ Trajectory: STABLE</div>" 
        if "IMPROVING" in text_out.upper():
            badge_html = "<div class='improving'>📈 Trajectory: IMPROVING</div>"
        elif "DECLINING" in text_out.upper():
            badge_html = "<div class='declining'>📉 Trajectory: DECLINING</div>"
            
        progress(0.8, desc="Annotating image...")
        final_present_img = present_img
        if "Deep Forensic" in audit_level:
            final_present_img = draw_forensic_targets(present_img, text_out)
            
        progress(1.0, desc="Audit Finalized.")
        return (past_img, final_present_img), badge_html, text_out
            
    except Exception as e:
        return None, "<div class='declining'>Analysis Failed</div>", str(e)


# --- UI LAYOUT ---
with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
    gr.Markdown("# 🏠 GroundTruth AI: Spatial Property Sentinel")
    
    # --- NEW: Address Search Bar ---
    with gr.Row():
        address_input = gr.Textbox(label="Property Address", placeholder="e.g., 1600 Pennsylvania Avenue NW, Washington, DC", scale=4)
        fetch_btn = gr.Button("🌍 Fetch Current Street View", variant="secondary", scale=1)

    # --- TOP ROW: The "Input Console" ---
    with gr.Row():
        with gr.Column(scale=1):
            p_img = gr.Image(label="Past Condition (Manual Upload)", type="pil")
        with gr.Column(scale=1):
            c_img = gr.Image(label="Current Condition (Auto-Fetched)", type="pil")
        with gr.Column(scale=1):
            audit_mode = gr.Radio(
                choices=["Standard Audit", "Deep Forensic"],
                value="Standard Audit",
                label="Select Audit Precision"
            )
            submit = gr.Button("Analyze Property Trajectory", variant="primary", size="lg")
            
    # --- MIDDLE ROW: The "Hero Output" ---
    with gr.Row():
        with gr.Column(scale=1):
            status_badge = gr.HTML()
            slider_out = gr.ImageSlider(label="Spatial Delta Comparison", type="pil", height=600)
            
    # --- BOTTOM ROW: The "Forensic Data" ---
    with gr.Row():
        with gr.Column(scale=1):
            with gr.Accordion("View Detailed Forensic Data", open=False):
                report_out = gr.Markdown()

    # --- EVENTS ---
    # 1. Wire the Fetch Button to the Maps API and output to the Current Image box
    fetch_btn.click(
        fn=fetch_streetview,
        inputs=[address_input],
        outputs=[c_img]
    )

    # 2. Wire the Analyze Button to the Robotics-ER Engine
    submit.click(
        fn=ground_truth_engine, 
        inputs=[p_img, c_img, audit_mode], 
        outputs=[slider_out, status_badge, report_out],
        show_progress="minimal"
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)