pdf_processor.py

import json
from langchain.document_loaders import PyPDFLoader
from models import ExtractionResult, EvaluationResult
from llm import get_llm

llm = get_llm()

def extract_answers_from_pdf(pdf_path: str) -> ExtractionResult:
    loader = PyPDFLoader(pdf_path)
    pages = loader.load_and_split()
    all_page_content = "\n".join(page.page_content for page in pages)
    
    extraction_schema = ExtractionResult.model_json_schema()
    system_message = (
        "You are a document analysis tool that extracts the options and correct answers from the provided document content. "
        "The output must be a JSON object that strictly follows the schema: " + json.dumps(extraction_schema, indent=2)
    )
    user_message = (
        "Please extract the correct answers and options (A, B, C, D, E) from the following document content:\n\n"
        + all_page_content
    )
    prompt = system_message + "\n\n" + user_message
    
    response = llm.invoke(prompt, response_format={"type": "json_object"})
    result = ExtractionResult.model_validate_json(response.content)
    return result

def evaluate_student(answer_key: ExtractionResult, student: ExtractionResult) -> EvaluationResult:
    evaluation_schema = EvaluationResult.model_json_schema()
    system_message = (
        "You are an academic evaluation tool that compares the answer key with a student's answers. "
        "Calculate the total marks, grade, and percentage based on the provided JSON objects. "
        "The output must be a JSON object that strictly follows the schema: " + json.dumps(evaluation_schema, indent=2)
    )
    user_message = (
        "Answer Key JSON:\n" + json.dumps(answer_key.model_dump(), indent=2) + "\n\n"
        "Student Answer JSON:\n" + json.dumps(student.model_dump(), indent=2)
    )
    prompt = system_message + "\n\n" + user_message
    
    response = llm.invoke(prompt, response_format={"type": "json_object"})
    evaluation_result = EvaluationResult.model_validate_json(response.content)
    return evaluation_result