PaperBrain – Automated Handwritten Answer Script Evaluation

Overview

PaperBrain is an AI-powered system that automates the evaluation of handwritten answer sheets for educational institutions.

Traditional evaluation methods rely on manual checking — which is time-consuming, error-prone, and resource-intensive. Teachers often spend days sorting, grading, and handling physical scripts, leading to delays and inconsistencies in results.

PaperBrain solves this by combining computer vision, OCR, and semantic AI to perform fair, fast, and scalable evaluations. It automatically aligns scanned copies, reads handwritten answers, maps them to reference solutions, and generates accurate, bias-free grade reports.

---

Solution Overview

PaperBrain uses a multi-agent AI pipeline designed for high accuracy, modularity, and adaptability. Each agent in the system performs a specific task in the evaluation process:

1. Planning Agent – Orchestrates the workflow, coordinating all other agents.
2. Agent 1: Align – Detects distortions and performs automatic alignment of scanned copies using ORB detection and homography transformation.
3. Agent 2: Find – Identifies regions of interest (answer areas) using pixel differencing and contour detection.
4. Agent 3: Read – Extracts handwritten text from identified regions using OCR models such as Tesseract or EasyOCR.
5. Agent 4: Grade – Compares extracted text with reference answers using semantic similarity and context-aware grading.
6. Report Generation – Produces a digital grade report that includes total marks, individual question scores, and flagged anomalies for manual review when required.

---

System Architecture

graph TD
    A[INPUTS<br/>Question Paper<br/>Scanned Answer Script<br/>Reference Answer Key] --> B[PLANNING AGENT<br/>Task Orchestration]
    
    B --> C[ALIGN AGENT<br/>ORB + Homography]
    B --> D[FIND AGENT<br/>Contours & Region Detection]
    B --> E[READ AGENT<br/>OCR: Tesseract/EasyOCR]
    
    C --> F[GRADE AGENT<br/>Semantic Matching & Scoring]
    D --> F
    E --> F
    
    F --> G[OUTPUT REPORT<br/>• Question-wise Marks<br/>• Total Score<br/>• Flagged Responses]
    
    style A fill:#2d3748,stroke:#4a5568,color:#fff
    style B fill:#2b6cb0,stroke:#2c5282,color:#fff
    style C fill:#2f855a,stroke:#276749,color:#fff
    style D fill:#2f855a,stroke:#276749,color:#fff
    style E fill:#2f855a,stroke:#276749,color:#fff
    style F fill:#c05621,stroke:#9c4221,color:#fff
    style G fill:#5a67d8,stroke:#434190,color:#fff

Loading

Key Capabilities:

• Automatic alignment of distorted answer sheets
• Semantic understanding of handwritten answers
• Smart anomaly flagging for unclear or incomplete responses
• Works for both MCQs and descriptive answers

---

Team Members

• Sanjeev Krishna
• Somasekhara S
• Bharat K
• Shivsrijit Verma
• Gaurav Mahesh