Overview

Traditional evaluation of descriptive answers is time-consuming, inconsistent, and difficult to scale for large classes. Verity addresses this problem by combining multiple AI techniques to analyze answer quality, semantic correctness, relevance, and language clarity. The system supports real-world academic workflows, including handwritten answer sheets and structured re-evaluation, making it suitable for institutional use.

How it works

Students can submit answers either by typing directly into the platform or by uploading scanned or photographed answer sheets. For handwritten submissions, OCR is applied to extract text, followed by intelligent preprocessing to remove metadata and noise. The cleaned answers are then evaluated using a multi-algorithm AI engine that compares student responses with expected answers defined by instructors.

AI Evaluation Engine

Verity uses a ten-algorithm scoring approach instead of relying on simple keyword matching. It combines exact and partial matching, TF-IDF cosine similarity, semantic similarity using deep learning embeddings, sentiment analysis, grammar checking, coherence, relevance, and probabilistic classification. These scores are weighted and merged into a final mark, closely aligning with how a human evaluator assesses subjective answers.

Re-evaluation Workflow

The platform includes a complete re-evaluation mechanism. Students can raise re-evaluation requests with reasons, and instructors can approve or reject them through a dedicated dashboard. Upon approval, the system reprocesses the original uploaded answer sheet and re-evaluates it automatically, ensuring fairness while keeping educators in control.

Key Features

• AI-based evaluation of descriptive answers

• Support for both typed and handwritten submissions

• OCR with preprocessing for high accuracy

• Transparent score breakdowns per question

• Role-based access for admins, teachers, and students

• Structured and auditable re-evaluation workflow

Tech Stack

• Backend: Flask, Python

• AI/NLP: NLTK, Sentence Transformers, scikit-learn

• OCR: Tesseract, Pillow

• Database: SQLite

• Frontend: HTML, CSS, Bootstrap

Impact

Verity reduces evaluation time by up to 80–90% while maintaining high alignment with human grading. It improves transparency for students, consistency for instructors, and scalability for institutions handling large volumes of descriptive assessments.