| from PIL import Image |
| from typing import Dict, Any |
| from .base_agent import LLMAgent, AgentResponse |
|
|
|
|
| class TextExtractionAgent(LLMAgent): |
| name = "text_extraction_agent" |
| description = "Extracts and interprets text from PDF pages using OCR and intelligent text processing" |
|
|
| def build_prompt(self, pil_image: Image.Image) -> str: |
| width, height = pil_image.size |
| return f"""Analyze this PDF page image (size: {width}x{height}). |
| |
| What text can be extracted? Describe what you see: |
| 1. What language(s) is the text in? |
| 2. What is the general layout structure? |
| 3. Are there any tables, forms, or special elements? |
| 4. What is the quality of the text - is it clear or blurry? |
| |
| Be specific about the content and any challenges for text extraction.""" |
|
|
| def execute(self, pil_image: Image.Image, reasoning: str) -> Dict[str, Any]: |
| import pytesseract |
|
|
| try: |
| text = pytesseract.image_to_string(pil_image) |
| return { |
| "text": text, |
| "success": True, |
| "method": "tesseract_ocr", |
| "reasoning": reasoning, |
| "language_detected": self._detect_lang(text), |
| } |
| except Exception as e: |
| return { |
| "text": "", |
| "success": False, |
| "error": str(e), |
| "reasoning": reasoning, |
| } |
|
|
| def _detect_lang(self, text: str) -> str: |
| if not text.strip(): |
| return "unknown" |
| return "en" |
|
|
