perception_agent.py

# ./perception_agent.py

import os
import pandas as pd
from docx import Document
from pypdf import PdfReader  # Cleanly leverages your requirements.txt package
from groq import Groq
from agent_logging import log_agent_action

#model = "llama-3.2-11b-vision-preview" # -> Decommissioned
model = "meta-llama/llama-4-scout-17b-16e-instruct"
client = Groq(api_key=os.getenv("GROQ_API_KEY"))

def read_image_file(file_path):
    """Uses Groq Vision capability to interpret images (.png, .jpg, .bmp)"""
    import base64
    try:
        log_agent_action("PERCEPTION", f"Encoding image for Vision API: {file_path}")
        with open(file_path, "rb") as image_file:
            encoded_string = base64.b64encode(image_file.read()).decode('utf-8')
        
        ext = os.path.splitext(file_path)[1].lower().replace(".", "")
        mime_type = f"image/{ext}" if ext != "jpg" else "image/jpeg"

        response = client.chat.completions.create(
            model=model,
            messages=[
                {
                    "role": "user",
                    "content": [
                        {"type": "text", "text": "Analyze this technical image. Extract all code, data tables, structural diagrams, or text precisely."},
                        {
                            "type": "image_url",
                            "image_url": {
                                "url": f"data:{mime_type};base64,{encoded_string}"
                            }
                        }
                    ]
                }
            ],
            temperature=0.2
        )
        log_agent_action("PERCEPTION_SUCCESS", f"Vision extraction complete for {file_path}")
        return f"\n--- Visual Content Extraction from {os.path.basename(file_path)} ---\n{response.choices[0].message.content}\n"
    except Exception as e:
        log_agent_action("PERCEPTION_ERROR", f"Vision interpretation failed: {str(e)}")
        return f"\n[Vision Error processing image {os.path.basename(file_path)}: {str(e)}]\n"

def read_document_file(file_path):
    """Universal router parsing text, code, spreadsheets, PDFs, and document assets"""
    ext = os.path.splitext(file_path)[1].lower()
    filename = os.path.basename(file_path)
    
    try:
        # 1. Plain Text and Markdown Layouts
        if ext in ['.txt', '.md', '.py', '.json', '.yaml', '.toml', '.css', '.html']:
            log_agent_action("PERCEPTION", f"Reading plaintext structure: {filename}")
            with open(file_path, "r", encoding="utf-8", errors="ignore") as f:
                return f"\n--- Content of File: {filename} ---\n{f.read()}\n"
        
        # 2. Excel Data Configurations
        elif ext in ['.xlsx', '.xls']:
            log_agent_action("PERCEPTION", f"Parsing Data Spreadsheet: {filename}")
            excel_data = pd.read_excel(file_path, sheet_name=None)
            combined_text = f"\n--- Spreadsheet Matrix Extraction: {filename} ---\n"
            for sheet_name, df in excel_data.items():
                combined_text += f"\nSheet: {sheet_name}\n"
                combined_text += df.to_markdown(index=False) + "\n"
            return combined_text
        
        # 3. Microsoft Word Processing
        elif ext == '.docx':
            log_agent_action("PERCEPTION", f"Extracting structural Word paragraphs: {filename}")
            doc = Document(file_path)
            paragraphs = [p.text for p in doc.paragraphs]
            return f"\n--- Document Text Extraction: {filename} ---\n" + "\n".join(paragraphs) + "\n"
        
        # 4. Portable Document Format (.pdf) Ingestion
        elif ext == '.pdf':
            log_agent_action("PERCEPTION", f"Initializing pypdf reader pipeline: {filename}")
            reader = PdfReader(file_path)
            pdf_text_buffer = []
            
            for index, page in enumerate(reader.pages):
                extracted_page_text = page.extract_text()
                if extracted_page_text:
                    pdf_text_buffer.append(f"--- Page {index + 1} ---\n{extracted_page_text}")
            
            if not pdf_text_buffer:
                log_agent_action("PERCEPTION_WARN", f"PDF contained no raw text layers (possible raw scan): {filename}")
                return f"\n[System Warning: '{filename}' appears to be an un-OCRed scanned image PDF. Please extract its pages as raw images for CoderG's Vision layer.]\n"
                
            log_agent_action("PERCEPTION_SUCCESS", f"Successfully parsed {len(pdf_text_buffer)} pages from {filename}")
            return f"\n--- PDF Document Content Ingestion: {filename} ---\n" + "\n".join(pdf_text_buffer) + "\n"
        
        # 5. Image Vector/Raster Formats
        elif ext in ['.png', '.jpg', '.jpeg', '.bmp']:
            return read_image_file(file_path)
            
        elif ext == '.doc':
            log_agent_action("PERCEPTION_WARN", f"Legacy format encountered: {filename}")
            return f"\n[System Error: Legacy format '{ext}' detected. Please convert '{filename}' to '.docx' for automated ingestion.]\n"
            
        else:
            log_agent_action("PERCEPTION_WARN", f"Unknown asset extension skipped: {filename}")
            return f"\n[System Warning: Unsupported file format '{ext}' for file '{filename}'. Skipping content ingestion.]\n"
            
    except Exception as e:
        log_agent_action("PERCEPTION_ERROR", f"Failed parsing {filename}: {str(e)}")
        return f"\n[Error processing document asset {filename}: {str(e)}]\n"