fraud_analyzer.py

import os
from google import genai
from PIL import Image
import io
import puremagic
from pypdf import PdfReader
import fitz  # PyMuPDF
from datetime import datetime

class FraudAnalyzer:
    def __init__(self, api_key):
        self.client = genai.Client(api_key=api_key)
        # LLM Services 3 Flash for reasoning and extraction
        self.model_flash = 'gemini-3-flash-preview' 
        # Nano Banana for specific image-centric analysis
        self.model_nano = 'nano-banana-pro-preview'

    def analyze_document(self, file_path):
        """Perform visual and structural analysis + data extraction."""
        file_ext = os.path.splitext(file_path)[1].lower()
        metadata = self._extract_metadata(file_path)
        
        current_datetime = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        visual_prompt = f"Today is {current_datetime}. Examine this image for visual manipulations: clones, font mismatches, edges around text, or retouching. Be extremely critical."
        extraction_prompt = f"""
        Today is {current_datetime}. Extract all text data from the following analysis and metadata. 
        Return a structured analysis in JSON format with:
        - label: A short descriptive name for the document.
        - amount: Any monetary amount found (default 0 or null).
        - fraud_score: An integer score from 0 to 100 indicating fraud probability (0 = clean, 100 = definitely fraud).
        - detailed_analysis: Your full reasoning.
        - extracted_fields: Key names, dates, and ID numbers.
        """

        image_input = None
        text_content = ""

        if file_ext in ['.png', '.jpg', '.jpeg']:
            image_input = Image.open(file_path)
        elif file_ext == '.pdf':
            # Extract text
            reader = PdfReader(file_path)
            for page in reader.pages:
                text_content += page.extract_text()
            
            # Convert first page to image for Nano Banana visual analysis
            try:
                doc = fitz.open(file_path)
                page = doc.load_page(0)
                pix = page.get_pixmap()
                img_data = pix.tobytes("png")
                image_input = Image.open(io.BytesIO(img_data))
                doc.close()
            except Exception as e:
                print(f"Error converting PDF to image: {e}")

        if image_input:
            # Dual analysis for images (or PDF converted to image)
            visual_resp = self.client.models.generate_content(
                model=self.model_nano,
                contents=[visual_prompt, image_input]
            )
            
            # Flash for Extraction & Combined Reasoning
            content_to_flash = [
                extraction_prompt, 
                f"Visual Analysis Report (from Nano Banana): {visual_resp.text}",
                f"Metadata: {metadata}"
            ]
            if text_content: # If PDF
                content_to_flash.append(f"Extracted Text: {text_content}")
            content_to_flash.append(image_input)
            
            combined_resp = self.client.models.generate_content(
                model=self.model_flash,
                contents=content_to_flash
            )
            
            return {
                "llm_analysis": f"Visual Report (Nano Banana):\n{visual_resp.text}\n\nFinal Reasoning (Flash):\n{combined_resp.text}",
                "metadata": metadata
            }
        else:
            # Fallback for text-only or unsupported image conversion
            combined_resp = self.client.models.generate_content(
                model=self.model_flash,
                contents=[extraction_prompt, f"Metadata: {metadata}", f"Text: {text_content}"]
            )
            return {
                "llm_analysis": f"Reasoning (Flash):\n{combined_resp.text}",
                "metadata": metadata
            }

    def _extract_metadata(self, file_path):
        """Extract file system and internal metadata."""
        try:
            file_mime = puremagic.from_file(file_path, mime=True)
        except Exception:
            file_mime = "application/octet-stream"
        
        stats = os.stat(file_path)
        metadata = {
            "mime_type": file_mime,
            "size_bytes": stats.st_size,
            "modified_time": stats.st_mtime
        }
        
        if "image" in file_mime:
            try:
                img = Image.open(file_path)
                exif = img._getexif()
                if exif:
                    metadata["exif"] = {str(k): str(v) for k, v in exif.items()}
            except Exception:
                pass
                
        return metadata