import re
import json
import os
from openai import OpenAI
from dotenv import load_dotenv

load_dotenv()

def get_structured_pan_data(ocr_text: str, app_id: str, provider: str = "gemini", logger=None) -> dict:
    """
    Use AI providers (OpenAI/Gemini) to extract Name, Father's Name, and DOB from PAN OCR text.
    The PAN number should be redacted BEFORE calling this function, but the prompt also enforces PII protection.
    """
    provider = provider.lower()
    if logger:
        logger.send_log_event(f"Application_id: {app_id}. In get_structured_pan_data. Provider: {provider}")
    
    config = {
        "gemini": {
            "api_key": os.getenv("GEMINI_API_KEY"),
            "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
            "model": "gemini-2.5-flash",
        },
        "openai": {
            "api_key": os.getenv("OPENAI_API_KEY"),
            "base_url": None,
            "model": "gpt-4o-mini",
        },
    }

    if provider not in config:
        if logger:
             logger.send_log_event(f"Application_id: {app_id}. Invalid provider '{provider}', falling back to gemini.")
        provider = "gemini"

    cfg = config[provider]
    
    if not cfg["api_key"]:
        if logger:
            logger.send_log_event(f"Application_id: {app_id}. API Key for {provider} missing.")
        return {"name": None, "father_name": None, "dob": None}

    client = OpenAI(api_key=cfg["api_key"], base_url=cfg.get("base_url"))

    system_prompt = (
        "You are an expert document data extractor for Indian PAN cards.\n"
        "You will receive OCR text from TWO engines: Tesseract (Engine 1) and EasyOCR (Engine 2).\n"
        "Tesseract may be faster but can be very noisy; EasyOCR is often more accurate for names.\n"
        "--- GOALS ---\n"
        "1. Extract: 'name' (Full Name), 'father_name' (Full Father's Name), 'dob' (DD/MM/YYYY).\n"
        "2. IMPORTANT: Do NOT truncate names. Capture all words.\n"
        "3. Cross-reference both engines to resolve typos or missing words.\n"
        "4. Return CLEAN, STANDARDIZED JSON only.\n"
        "5. IMPORTANT: DO NOT include the PAN number in your response.\n"
        "6. STRICT KEYWORD RULE: For 'father_name', only extract it if a corresponding label (e.g., 'Father', 'Father's Name', 'Pita') is detected in the OCR text. If no father-related keyword is found, return null for 'father_name' even if there is text present that looks like a name.\n"
        "Required keys: name, father_name, dob."
    )

    try:
        response = client.chat.completions.create(
            model=cfg["model"],
            temperature=0,
            messages=[
                {"role": "system", "content": system_prompt},
                {"role": "user", "content": f"OCR Text from PAN Card:\n{ocr_text}"}
            ],
        )

        content = response.choices[0].message.content
        # Remove markdown code fences
        if content.startswith("```"):
            content = content.split("```")[1]
            content = content.replace("json", "", 1).strip()
            
        if logger:
            logger.send_log_event(f"Application_id: {app_id}. AI Response: {content}")
        
        return json.loads(content)

    except Exception as e:
        if logger:
            logger.send_log_event(f"Application_id: {app_id}. Error in get_structured_pan_data: {str(e)}")
        return {"name": None, "father_name": None, "dob": None}