src/vision_processor.py

import os
import base64
from pdf2image import convert_from_path
from langchain_core.documents import Document
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage

class VisionProcessor:
    def __init__(self):
        self.vision_model = ChatOpenAI(model="gpt-4o", max_tokens=1024)
        
        # PRO FIX: Point to local Poppler bin
        # This assumes 'poppler' folder is in the project root
        self.poppler_path = os.path.join(os.getcwd(), "poppler", "Library", "bin")

    def process_visual_pdf(self, pdf_path):
        print(f"   👁️ Processing Visual PDF: {os.path.basename(pdf_path)}...")
        documents = []
        
        try:
            # Check if our local poppler exists
            if not os.path.exists(self.poppler_path):
                print(f"     ❌ Error: Poppler not found at {self.poppler_path}")
                return []

            # 1. Convert PDF pages to Images (Using local poppler)
            images = convert_from_path(pdf_path, fmt="jpeg", poppler_path=self.poppler_path)
            
            print(f"     -> Extracted {len(images)} images (pages) from PDF.")

            # 2. Analyze first 3 pages (Cost Saving Mode)
            for i, img in enumerate(images[:3]): 
                print(f"     -> Analyzing Page {i+1} with GPT-4o Vision...")
                
                # Base64 Encode
                import io
                buffered = io.BytesIO()
                img.save(buffered, format="JPEG")
                img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")

                # 3. Send to GPT-4o
                response = self.vision_model.invoke(
                    [
                        HumanMessage(
                            content=[
                                {"type": "text", "text": "Describe this image in detail. If it is a graph, extract the data points. If it is a table, transcribe it."},
                                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img_str}"}},
                            ]
                        )
                    ]
                )
                
                description = response.content
                
                doc = Document(
                    page_content=f"IMAGE DESCRIPTION (Page {i+1}): {description}",
                    metadata={
                        "source": os.path.basename(pdf_path),
                        "page": i+1,
                        "category": "visual_data"
                    }
                )
                documents.append(doc)

            if len(images) > 3:
                print("     ℹ️ Limited to first 3 pages for POC cost safety.")

        except Exception as e:
            print(f"     ❌ Vision Error: {e}")
        
        return documents