Exam-Grading-System

Exam-Grading-System is a lightweight, self‑contained exam grading tool. It allows educators to create exams, add questions and students, upload PDF answer sheets and receive automated scoring. The backend is implemented with FastAPI and SQLite using only packages available in this environment; the frontend is rendered with Jinja2 templates and styled with a minimalist stylesheet inspired by shadcn/ui's white palette.

Features

• Exam management – create exams with titles, subjects and instructions, add questions with ideal answers and point values and register students.
• Answer sheet upload – upload a PDF answer sheet for each student. The server extracts text from the PDF, splits it into responses and compares each response to the ideal answer using RapidFuzz's token set ratio. The total score and per‑question breakdown are stored.
• Results dashboard – view all students, their uploaded sheets and scores; drill down to see a side‑by‑side comparison of the student's answer and the canonical answer.
• CSV export – download a CSV file summarising scores for the entire class.

Running the application

1. Make sure you have Python 3.11 available. All required dependencies are included in this environment (FastAPI, uvicorn, PyMuPDF, RapidFuzz). You do not need to install Django; this project uses FastAPI because external package installation is restricted in this environment.

2. Create the database and start the web server:

   uvicorn src.main:app --host 0.0.0.0 --port 8000 --reload

   This will start a server on http://127.0.0.1:8000/. Navigate there in your browser to begin creating exams.

3. To stop the server press Ctrl+C in the terminal.

Integrating a real AI model

core login in src/utility/utility.py:

You have to use Useing two AI models

• Florence-2: For OCR and text extraction from scanned answer sheets

• LLaMA 3 8B Instruct: For evaluating and grading student answers against ideal answers

Core Functions

1. Text Extraction with Florence-2

* extract_with_paddleocr(image_path: str) -> str

* Converts PDF pages to images using pdf2image

* Processes each image with Florence-2 model

* Uses <OCR_WITH_REGION> task prompt for optimized text extraction

* Returns structured OCR data with bounding boxes and text labels

2. Key Features:

* Handles multi-page PDF documents

* Automatic cleanup of temporary image files

* Robust error handling for corrupted pages

* Returns both text content and spatial information

3. Answer Evaluation with LLaMA

* evaluate_answer(student_answer: str, ideal_answer: str, point_value: float) -> float

* Uses LLaMA 3 8B Instruct model via Replicate API

* Compares student answers against ideal answers

* Returns numerical score based on point value

* Includes strict prompt engineering for consistent grading

Grading Prompt Structure:

  prompt = f"""
    You are grading a short answer question. The ideal answer is:
    "{ideal_answer}"

    The student's answer is:
    "{student_answer}"

    Respond with a single number between 0 and {point_value} representing
    the score. Do not provide explanations, only the number.
  """

Workflow:

• PDF Processing: Converts PDF to individual page images

• Text Extraction: Uses Florence-2 to extract answers from each page

• Answer Matching: Aligns extracted answers with database questions

• AI Grading: Evaluates each answer using LLaMA model

• Results Storage: Saves scores to database with timestamps

Database Operations:

• Updates answers table with individual question scores

• Updates answer_sheets table with total score and evaluation timestamp

• Maintains referential integrity with student and exam data

Model Configuration

Florence-2 Setup Located in configuration/florence2_configuration.py

 processor = AutoProcessor.from_pretrained("microsoft/Florence-2-large", trust_remote_code=True)
 
 model = AutoModelForCausalLM.from_pretrained("microsoft/Florence-2-large", trust_remote_code=True, torch_dtype=torch.bfloat16)

LLaMA Configuration

Model: meta/meta-llama-3-8b-instruct

Access: Via Replicate API with secure token authentication

Output: Strict numerical scoring only

You will need to set your Replicate API token in the environment variable REPLICATE_API_TOKEN before running the server. See the Replicate documentation for details.

API Endpoints List

Dashboard & Navigation

 GET / - Home page listing all exams

 GET /exams/new - Create exam form

Exam Management

 POST /exams/new - Create new exam

 GET /exams/{exam_id} - Get exam details

 POST /exams/{exam_id}/questions - Add question to exam

 POST /exams/{exam_id}/students - Add student to exam

Answer Sheet Management

 POST /exams/{exam_id}/upload - Upload answer sheet PDF

 GET /answer_sheets/{sheet_id}/download - Download answer sheet PDF

 GET /answer_sheets/{sheet_id} - Get answer sheet details

Results & Reporting

  GET /exams/{exam_id}/results - View exam results

  GET /exams/{exam_id}/export - Export results as CSV