import pdfplumber import pandas as pd import re import os from getpass import getpass from src.logger import logger # Try importing OCR dependencies try: import pytesseract from pdf2image import convert_from_path OCR_AVAILABLE = True except ImportError: OCR_AVAILABLE = False logger.warning("OCR Dependencies (pytesseract/pdf2image) not found. Scanned PDFs cannot be processed.") # Delayed import to avoid circular dependency def get_classifier(): from src.classifier import tag_transactions return tag_transactions def parse_with_neural_engine(pdf_path): logger.info(f"🧠 NEURAL ENGINE: Scanning {pdf_path}...") transactions = [] pdf = None try: # 1. ATTEMPT OPENING try: pdf = pdfplumber.open(pdf_path) except Exception as e: if "password" in str(e).lower() or "authenticate" in str(e).lower(): logger.warning("File is encrypted.") user_pass = getpass(" [PROMPT] Enter PDF Password: ") pdf = pdfplumber.open(pdf_path, password=user_pass) else: logger.error(f"PDF Initialization Failed: {str(e)}") return None # 2. VALIDATE PAGES if not pdf.pages: logger.error("PDF contains no pages.") return None # 3. TEXT EXTRACTION (Hybrid: Text -> OCR) full_text = "" for i, page in enumerate(pdf.pages): text = page.extract_text() if text and len(text.strip()) > 50: full_text += text + "\n" else: logger.warning(f"Page {i+1} seems to be an image. Attempting OCR...") if OCR_AVAILABLE: try: # Convert specific page to image # Note: poppler_path might need to be configured on Windows, but on Linux usually auto-detected images = convert_from_path(pdf_path, first_page=i+1, last_page=i+1) for img in images: ocr_text = pytesseract.image_to_string(img) full_text += ocr_text + "\n" logger.info(f"OCR extracted {len(ocr_text)} chars from Page {i+1}") except Exception as ocr_e: logger.error(f"OCR Failed for Page {i+1}: {ocr_e}") logger.info("Ensure 'poppler-utils' and 'tesseract-ocr' are installed.") else: logger.error("OCR is not available (dependencies missing). Skipping page.") if not full_text.strip(): logger.error("No extractable text found. If this is a scanned photo, ensure OCR is set up.") return None lines = full_text.split('\n') logger.info(f"Analyzing {len(lines)} text lines...") # 4. PATTERN RECOGNITION (REGEX) # Enhanced to be more robust for OCR noise date_pattern = re.compile(r'\b(\d{1,2}[-/\s][A-Za-z]{3}[-/\s]\d{2,4})\b') money_pattern = re.compile(r'[\d,]+\.\d{2}') # Matches 12,345.00 current_txn = {} for line in lines: # Cleanup noise cleaned_line = line.strip() if not cleaned_line: continue match = date_pattern.search(cleaned_line) if match: if current_txn: transactions.append(current_txn) date_str = match.group(0) remaining = cleaned_line.replace(date_str, "") money = money_pattern.findall(remaining) val = 0.0 if money: # Heuristic: Usually the last money amount is the balance, second last is transaction amount # But in statements, sometimes it's Credit | Debit | Balance. # This logic is simplistic but works for many standard formats. # OCR might introduce commas or dots wrongly. amt_str = money[-1] if len(money) == 1 else money[-2] if len(money) > 1 else "0" try: val = float(amt_str.replace(',', '')) except: val = 0.0 # Determine Dr/Cr is_deposit = any(x in cleaned_line.lower() for x in ["cr", "credit", "deposit"]) current_txn = { "Date": date_str, "Description": re.sub(r'\s+', ' ', remaining).strip(), "Withdrawal": val if not is_deposit else 0.0, "Deposit": val if is_deposit else 0.0 } elif current_txn: current_txn['Description'] += " " + cleaned_line if current_txn: transactions.append(current_txn) df = pd.DataFrame(transactions) if not df.empty: logger.info(f"SUCCESS: Reconstructed {len(df)} transactions.") # Real-time categorization try: tagger = get_classifier() df = tagger(df) logger.info("🧠 NEURAL ENGINE: Auto-categorized transactions via AI.") except Exception as e: logger.warning(f"AI Categorization during parsing failed: {e}") return df else: logger.error("Logic failed to find dates in text.") return None except Exception as e: logger.critical(f"System Crash in Parser: {str(e)}") return None finally: if pdf: pdf.close()