yeswanthvarma/answer-evaluation-app

Answer Evaluation App ✍️

📝 Handwritten Answer Evaluation App using OCR + XLNet

An end-to-end machine learning application that extracts handwritten answers from images and evaluates them using a custom XLNet model trained on semantic similarity.

🚀 Live Demo

👉 Try it on Hugging Face Spaces

📌 Project Overview

This app takes images of handwritten answers and performs:

1. 1.OCR to extract text from question, student answer, and reference answer.
2. 2.Similarity scoring using a custom-trained XLNet model.
3. 3.Bonus logic to adjust the final score based on thresholds.
4. 4.User interface to upload images and view the evaluated score.

🧠 Core Technologies

• —FastAPI: Web framework
• —EasyOCR: For extracting handwritten text
• —Hugging Face Transformers: XLNet model
• —Custom Training: Trained on Q-A-R triplets
• —Docker: For containerized deployment
• —Hugging Face Spaces: Live hosted app

📦 Folder Structure

answer-evaluation-app/
├── app.py                     # FastAPI application
├── requirements.txt           # Dependencies
├── Dockerfile                 # For Hugging Face deployment
├── utils/
│   ├── image_processor.py     # EasyOCR + preprocessing
│   └── xlnet_model.py         # Model load and prediction
├── templates/
│   └── index.html             # Frontend HTML
├── static/
│   ├── css/style.css          # UI styling
│   ├── js/main.js             # JS for client interaction
│   └── uploads/               # Uploaded image storage

🔍 Model Details

• —Base Model: xlnet-base-cased (Hugging Face)
• —Custom Trained On: Question, student answer, reference answer, and human-evaluated scores
• —Loss: MSELoss
• —Output: Score from 0 to 100 ---

✍️ Sample Use Case

• —Upload 3 images:
• —Question image
• —Student handwritten answer
• —Reference answer
• —App will:
• —Extract text
• —Score similarity using model
• —Apply bonus logic
• —Display final score and extracted text

🧑‍💻 Author

Yeswanth Varma Gottumukkala

• —Email: yeswanthvarma.g@gmail.com

📜 License

This project is for educational and research purposes. Model and app are freely available to explore on Hugging Face Spaces.