Create pdf_processor.py
Browse files- pdf_processor.py +50 -0
pdf_processor.py
ADDED
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from langchain.document_loaders import PyPDFLoader
|
3 |
+
from models import ExtractionResult, EvaluationResult
|
4 |
+
from llm import get_llm
|
5 |
+
|
6 |
+
llm = get_llm()
|
7 |
+
|
8 |
+
def extract_answers_from_pdf(pdf_path: str) -> ExtractionResult:
|
9 |
+
"""
|
10 |
+
Loads a PDF, extracts its content, and uses the LLM to output a JSON of the answers.
|
11 |
+
"""
|
12 |
+
loader = PyPDFLoader(pdf_path)
|
13 |
+
pages = loader.load_and_split()
|
14 |
+
all_page_content = "\n".join(page.page_content for page in pages)
|
15 |
+
|
16 |
+
# Build the system message with JSON schema instructions.
|
17 |
+
extraction_schema = ExtractionResult.model_json_schema()
|
18 |
+
system_message = (
|
19 |
+
"You are a document analysis tool that extracts the options and correct answers from the provided document content. "
|
20 |
+
"The output must be a JSON object that strictly follows the schema: " + json.dumps(extraction_schema, indent=2)
|
21 |
+
)
|
22 |
+
user_message = (
|
23 |
+
"Please extract the correct answers and options (A, B, C, D, E) from the following document content:\n\n"
|
24 |
+
+ all_page_content
|
25 |
+
)
|
26 |
+
prompt = system_message + "\n\n" + user_message
|
27 |
+
|
28 |
+
response = llm.invoke(prompt, response_format={"type": "json_object"})
|
29 |
+
result = ExtractionResult.model_validate_json(response.content)
|
30 |
+
return result
|
31 |
+
|
32 |
+
def evaluate_student(answer_key: ExtractionResult, student: ExtractionResult) -> EvaluationResult:
|
33 |
+
"""
|
34 |
+
Compares the answer key with a student's answers and returns the evaluation result.
|
35 |
+
"""
|
36 |
+
evaluation_schema = EvaluationResult.model_json_schema()
|
37 |
+
system_message = (
|
38 |
+
"You are an academic evaluation tool that compares the answer key with a student's answers. "
|
39 |
+
"Calculate the total marks, grade, and percentage based on the provided JSON objects. "
|
40 |
+
"The output must be a JSON object that strictly follows the schema: " + json.dumps(evaluation_schema, indent=2)
|
41 |
+
)
|
42 |
+
user_message = (
|
43 |
+
"Answer Key JSON:\n" + json.dumps(answer_key.model_dump(), indent=2) + "\n\n"
|
44 |
+
"Student Answer JSON:\n" + json.dumps(student.model_dump(), indent=2)
|
45 |
+
)
|
46 |
+
prompt = system_message + "\n\n" + user_message
|
47 |
+
|
48 |
+
response = llm.invoke(prompt, response_format={"type": "json_object"})
|
49 |
+
evaluation_result = EvaluationResult.model_validate_json(response.content)
|
50 |
+
return evaluation_result
|