app.py

from flask import Flask, request, jsonify
import os
import requests
from PIL import Image
import tempfile
from PyPDF2 import PdfReader
from threading import Thread
import io
import fitz
from groq import Groq
from queue import Queue
import base64

# Initialize Flask app
app = Flask(__name__)

# Get API tokens from environment variables
HF_TOKEN = os.environ.get('HF_TOKEN')
GROQ_API_KEY = os.environ.get('GROQ_API_KEY')

# Initialize Groq client
client = Groq(api_key=GROQ_API_KEY)

# Configuration for low memory mode (maintaining original functionality)
LOW_MEMORY = os.getenv("LOW_MEMORY", "0") == "1"
print(f"Low memory mode: {LOW_MEMORY}")

class TextStreamer:
    def __init__(self):
        self.queue = Queue()
        self.buffer = ""
        
    def put(self, text):
        self.queue.put(text)
        
    def __iter__(self):
        while True:
            if not self.queue.empty():
                text = self.queue.get()
                if text is None:  # End signal
                    break
                yield text
            else:
                continue

def extract_image_from_pdf(pdf_url, dpi=75):
    """
    Extract first page of PDF as image in memory
    
    Args:
        pdf_url (str): URL of PDF
        dpi (int): Image resolution
    
    Returns:
        PIL.Image: First page as image or None
    """
    try:
        # Download PDF
        print(f"Attempting to download PDF from: {pdf_url}")
        # Download PDF
        response = requests.get(pdf_url, timeout=30)
        response.raise_for_status()
        print(f"PDF download status code: {response.status_code}")
        
        # Open PDF from bytes
        print("Opening PDF document...")
        pdf_document = fitz.open(stream=response.content, filetype="pdf")
        
        # Get first page
        print("Getting first page...")
        first_page = pdf_document[0]
        
        # Render page to pixmap
        print("Rendering page to pixmap...")
        pix = first_page.get_pixmap(matrix=fitz.Matrix(dpi/72, dpi/72))
        
        # Convert to PIL Image
        print("Converting to PIL Image...")
        img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
        
        pdf_document.close()
        print("Successfully extracted image from PDF")
        return img
    
    except Exception as e:
        print(f"Error extracting first page: {e}")
        return None

def predict_image(image_url, text, file_pref):
    try:
        if file_pref == 'img':
            response = requests.get(image_url)
            response.raise_for_status()
            image = Image.open(io.BytesIO(response.content)).convert("RGB")
        else: 
            print("Extracting image from PDF...")
            image = extract_image_from_pdf(image_url)
            
            # Convert the image to base64
            buffered = io.BytesIO()
            image.save(buffered, format="PNG")
            img_str = base64.b64encode(buffered.getvalue()).decode()
            image_url = f"data:image/png;base64,{img_str}"
            
        print(f"Image URL being sent to Groq: {image_url[:100]}...") 
            
        streamer = TextStreamer()
        
        def generate_response():
            try:
                completion = client.chat.completions.create(
                    model="llama-3.2-11b-vision-preview",
                    messages=[
                        {
                            "role": "user",
                            "content": [
                                {
                                    "type": "image_url",
                                    "image_url": {"url": image_url}
                                },
                                {
                                    "type": "text",
                                    "text": text
                                }
                            ]
                        }
                    ],
                    temperature=0.7,
                    max_tokens=4096,
                    top_p=1,
                    stream=True

                    # temperature=1,
                    # max_completion_tokens=4096,
                    # top_p=1,
                    # stream=False,
                    # stop=None,
                )
                print(f"Completions: {completion}")
                for chunk in completion:
                    if chunk.choices[0].delta.content:
                        streamer.put(chunk.choices[0].delta.content)
                streamer.put(None)  # Signal the end
            except Exception as e:
                print(f"Error in generate_response: {e}")
                streamer.put(None)

        thread = Thread(target=generate_response)
        thread.start()
        
        buffer = ""
        for new_text in streamer:
            buffer += new_text
            # buffer = completion.choices[0].message
            # print(buffer)
        return buffer
        
    except Exception as e:
        raise ValueError(f"Error during prediction: {str(e)}")

def extract_text_from_pdf(pdf_url):
    try:
        response = requests.get(pdf_url)
        response.raise_for_status()
        with tempfile.NamedTemporaryFile(delete=False) as temp_pdf:
            temp_pdf.write(response.content)
            temp_pdf_path = temp_pdf.name
        
        reader = PdfReader(temp_pdf_path)
        text = ""
        for page in reader.pages:
            text += page.extract_text()
        
        os.remove(temp_pdf_path)
        return text
    except Exception as e:
        raise ValueError(f"Error extracting text from PDF: {str(e)}")

def predict_text(text):
    streamer = TextStreamer()
    
    def generate_response():
        try:
            completion = client.chat.completions.create(
                model="meta-llama/llama-4-scout-17b-16e-instruct",
                messages=[
                    {
                        "role": "user",
                        "content": text
                    }
                ],
                temperature=0.7,
                max_tokens=4096,
                top_p=1,
                stream=True
            )
            
            for chunk in completion:
                if chunk.choices[0].delta.content:
                    streamer.put(chunk.choices[0].delta.content)
            streamer.put(None)  # Signal the end
        except Exception as e:
            print(f"Error in generate_response: {e}")
            streamer.put(None)

    thread = Thread(target=generate_response)
    thread.start()
    
    buffer = ""
    for new_text in streamer:
        buffer += new_text
        
    return buffer

# [Rest of the prompts remain exactly the same as in original]
PROMPT = (
    "Extract the following information as per this format:\n"
    "'Course Code:'\n"
    "'Course Name:'\n"
    "'Course Description:'\n"
    "'Course Credits:'\n"
    "'Course Learning Outcomes:'\n"
    "'Delivery Method:'\n"
    "'Prerequisite(s):'\n"
    "'Co-requisite(s):'\n"
    "'Materials:'\n"
    "'Topical Outline:'\n"
    "Do not add anything else except the required information from this text."
)

PROMPT_SKILLS = (
    "Provide skills based on the Lightcast Open Skills Taxonomy in categories as:\n"
    "'Primary Skills' (the degree program or certification),\n"
    "'Secondary Skills', and\n"
    "'Tertiary Skills'."
)

PROMPT_IMAGE_STUDENT = (
    "You are a highly intelligent assistant designed to analyze images and extract structured information from them. "
    "Your task is to analyze the given image of a student's academic record and generate a response in the exact JSON format provided below. "
    "If any specific information is missing or unavailable in the image, replace the corresponding field with null. "
    "Ensure the format is consistent, strictly adhering to the structure shown below.\n\n"
    "Required JSON Format:\n\n"
    "{\n"
    '  "student": {\n'
    '    "name": "string",\n'
    '    "id": "string",\n'
    '    "dob": "string",\n'
    '    "original_start_date": "string",\n'
    '    "cumulative_gpa": "string",\n'
    '    "program": "string",\n'
    '    "status": "string"\n'
    '  }\n'
    "}\n\n"
    "Instructions:\n\n"
    "1. Extract the student's general information as displayed in the image.\n"
    "2. Use null for any missing or unavailable information.\n"
    "3. Format the extracted data exactly as shown above. Do not deviate from this structure.\n"
    "4. Ensure accurate field names and proper nesting.\n"
    "5. Return only the 'student' section as JSON.\n"
)

PROMPT_IMAGE_COURSES = (
    "You are a highly intelligent assistant designed to analyze images and extract structured information from them. "
    "Your task is to analyze the given image of a student's academic record and generate a response in the exact JSON format provided below. "
    "If any specific information is missing or unavailable in the image, replace the corresponding field with null. "
    "Ensure the format is consistent, strictly adhering to the structure shown below.\n\n"
    "Required JSON Format:\n\n"
    "{\n"
    '  "courses": [\n'
    '    {\n'
    '      "transfer_institution": "string",\n'
    '      "course_code": "string",\n'
    '      "course_name": "string",\n'
    '      "credits_attempted": number,\n'
    '      "credits_earned": number,\n'
    '      "grade": "string",\n'
    '      "quality_points": number,\n'
    '      "semester_code": "string",\n'
    '      "semester_dates": "string"\n'
    '    }\n'
    "    // Additional courses can be added here\n"
    "  ]\n"
    "}\n\n"
    "Instructions:\n\n"
    "1. Extract the course details as displayed in the image.\n"
    "2. Use null for any missing or unavailable information.\n"
    "3. Format the extracted data exactly as shown above. Do not deviate from this structure.\n"
    "4. Ensure accurate field names and proper nesting.\n"
    "5. Return only the 'courses' section as JSON.\n"
)

@app.route("/", methods=["GET"])
def home():
    return jsonify({"message": "Welcome to the PDF Extraction API. Use the /extract endpoint to extract information."})

@app.route("/favicon.ico")
def favicon():
    return "", 204

@app.route("/extract", methods=["POST"])
def extract_info():
    data = request.json
    if not data:
        return jsonify({"error": "Please provide a PDF URL in the request body."}), 400
    
    try:
        if data["url"] is not None:
            pdf_url = data["url"]
            pdf_text = extract_text_from_pdf(pdf_url)
            prompt = f"{PROMPT}\n\n{pdf_text}"
            response = predict_text(prompt)
        else:
            response = ''
            
        if data["skills"] == True:
            if response: 
                prompt_skills = f"{PROMPT_SKILLS} using this information only -- {response}"
                response_skills = predict_text(prompt_skills)
            else:
                response_skills = ''
        else:
            response_skills = ''
            
        if data["img_url"] is not None:
            prompt_student = f"{PROMPT_IMAGE_STUDENT}\n"
            prompt_courses = f"{PROMPT_IMAGE_COURSES}\n"
            img_url = data["img_url"]
            file_pref = data["file_pref"]
            response_student = predict_image(img_url, prompt_student, file_pref)
            response_courses = predict_image(img_url, prompt_courses, file_pref)
            response_image = response_student + response_courses 
        else:
            response_image = ''
            
        return jsonify({"extracted_info": response + "\n" + response_skills + "\n" + response_image})

    except Exception as e:
        return jsonify({"error": str(e)}), 500

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860)