📝 AutoGradeAI

Intelligent Automated Grading using OCR and Semantic Similarity

A smart AI-powered grading system that evaluates handwritten student answers by combining OCR, NLP, and semantic similarity. Upload an answer sheet, compare it with the model answer, and automatically compute a fair grade.

---

📖 What is AutoGradeAI?

AutoGradeAI is an AI-powered automated grading system designed to streamline the evaluation of handwritten student answers.

Traditional grading requires teachers to manually read and score each answer. This project eliminates that burden by combining OCR (Optical Character Recognition) and semantic similarity models to evaluate answers intelligently.

When an answer sheet is uploaded:

1. EasyOCR extracts text from the scanned handwritten answer.
2. TextBlob corrects spelling errors introduced during OCR extraction.
3. SentenceTransformers generates embeddings for both student and model answers.
4. The system computes a semantic similarity score between them.
5. The similarity score is converted into a final grade based on configurable weightage.

The result is a system that evaluates answers based on meaning rather than exact wording, making grading more fair and efficient.

---

✨ Features

Feature	Description
🧠 AI-Powered Grading	Uses semantic similarity instead of simple keyword matching
📝 Handwritten Text Extraction	EasyOCR extracts text directly from scanned answer sheets
✍️ Spelling Correction	TextBlob automatically corrects OCR mistakes
🎯 Semantic Similarity Evaluation	SentenceTransformers compares meaning between answers
⚖️ Custom Weightage System	Teachers can define marks per question
📊 Similarity-Based Grading	Marks are calculated from similarity score percentages
🌐 Interactive Web App	Built with Streamlit for easy use

---

🏗️ System Architecture

Grading Pipeline

graph TD
    A[👤 Teacher uploads answer sheet] --> B[Streamlit Web Interface]
    B --> C[Image Processing]
    C --> D[EasyOCR Text Extraction]
    D --> E[TextBlob Spelling Correction]

    E --> F[Student Answer Embedding]
    G[Model Answer] --> H[Model Embedding]

    F --> I[Semantic Similarity Calculation]
    H --> I

    I --> J[Similarity Score %]
    J --> K[Weightage-based Grade Calculation]
    K --> L[Final Marks Output]

Loading

---

Processing Workflow

The grading process consists of four main stages:

1️⃣ OCR Extraction

• EasyOCR scans the uploaded answer sheet
• Extracts the handwritten content into machine-readable text

2️⃣ Text Cleaning

• TextBlob performs spelling correction
• Ensures OCR errors do not affect evaluation

3️⃣ Semantic Comparison

• SentenceTransformers generates embeddings
• Cosine similarity measures how closely answers match

4️⃣ Grade Computation

• Similarity score converted into marks
• Adjusted based on teacher-defined weightage

---

🛠️ Technology Stack

Core System

Component	Technology
Programming Language	Python
Web Interface	Streamlit
OCR Engine	EasyOCR
Semantic Similarity	SentenceTransformers
Spelling Correction	TextBlob
Image Processing	PIL
Numerical Processing	NumPy

---

📂 Project Structure

automated-grading-system/
│
├── app.py                 # Streamlit application
├── grading_engine.py      # Core grading logic
├── ocr_processor.py       # EasyOCR extraction
├── similarity_model.py    # SentenceTransformer embedding + similarity
├── utils/
│   ├── spell_correct.py   # TextBlob spelling correction
│   └── image_utils.py     # Image preprocessing utilities
│
├── requirements.txt       # Python dependencies
└── README.md

---

🚀 Installation & Setup

Prerequisites

• Python 3.9+
• pip
• GPU optional (for faster OCR processing)

---

1️⃣ Clone the Repository

git clone https://github.com/your-username/automated-grading-system.git
cd automated-grading-system

---

2️⃣ Install Dependencies

pip install -r requirements.txt

This installs:

• Streamlit
• EasyOCR
• SentenceTransformers
• TextBlob
• NumPy
• Pillow

---

🏃 Running the Application

Launch the Streamlit app:

streamlit run app.py

The application will start locally:

http://localhost:8501

---

🌐 Application Workflow

Step 1 — Upload Answer Sheet

Upload a scanned image of the student answer sheet.

Supported formats:

• PNG
• JPG
• JPEG

---

Step 2 — Enter Model Answer

Provide the correct reference answer that will be used for evaluation.

---

Step 3 — Assign Weightage

Define the maximum marks for the question.

Example:

Total marks: 10

---

Step 4 — Evaluate

Click Evaluate, and the system will:

• Extract text using OCR
• Correct spelling
• Compute semantic similarity
• Calculate final marks

---

📊 Example Output

Metric	Result
Corrected Student Answer	Displayed in UI
Similarity Score	87%
Total Marks	10
Marks Awarded	8.7

---

⚙️ Configuration

Setting	File	Description
OCR Model	ocr_processor.py	EasyOCR language settings
Similarity Model	similarity_model.py	SentenceTransformer model
Grading Logic	grading_engine.py	Score calculation formula
UI Interface	app.py	Streamlit UI

---

🐛 Known Issues & Troubleshooting

OCR accuracy issues

Improve results by:

• Using high-resolution images
• Avoiding skewed answer sheets
• Ensuring clear handwriting

---

Slow processing

OCR and embeddings can be slow on CPU.

Solutions:

• Enable GPU for EasyOCR
• Use lighter SentenceTransformer models

---

🔮 Future Improvements

• 🌍 Multi-language grading
• ✍️ Improved handwriting recognition
• 🧠 Context-aware answer evaluation
• 🗂️ Student result database
• 📊 Analytics dashboard for teachers
• 📑 Bulk grading support

---

📑 License

This project is licensed under the MIT License.

---

🙏 Acknowledgements

This project was built using several powerful open-source technologies:

• EasyOCR — text extraction from images
• TextBlob — spelling correction and NLP utilities
• SentenceTransformers — semantic similarity modeling
• Streamlit — rapid web app development for AI projects

---

Transforming traditional grading into intelligent AI evaluation.

AutoGradeAI — because grading should measure understanding, not just keywords.