CopyChecker-AI

AI-Powered Intelligent Answer Evaluation System

A full-stack AI system that performs automated evaluation of student answer sheets based upon model answer key provided.

---

🌐 Overview

CopyChecker-AI is an intelligent academic evaluation system that:

• 📄 Accepts Model Answer Key and Student Answer PDFs
• 🧠 Extracts text and images using advanced OCR pipelines
• 🖼 Performs layout-aware segmentation using YOLOv8
• 🔍 Detects structural and semantic similarities
• 🤖 Uses LLM reasoning (Groq API) for evaluation scoring
• ⚡ Generates structured evaluation results

The system goes beyond simple text matching by combining:

• Computer Vision
• OCR
• Document Layout Analysis
• Large Language Model evaluation
• Parallel processing pipelines

This makes it significantly more advanced than traditional plagiarism detection tools.

---

🎯 Problem Statement

Traditional answer checkers:

• ❌ Rely only on text matching
• ❌ Fail on scanned handwritten PDFs
• ❌ Ignore layout structure
• ❌ Cannot evaluate answer quality

CopyChecker-AI solves this by:

• Converting PDFs into high-resolution images
• Performing layout-aware segmentation
• Extracting text via OCR
• Using LLM-based semantic comparison
• Producing contextual evaluation output

---

🧠 Core Technologies Used

🔹 Backend & Frontend

• Python
• Flask (Backend interface)
• HTML & CSS (Frontend interface)

🔹 Computer Vision

• OpenCV
• YOLOv8 (Ultralytics)
• Image preprocessing with CLAHE
• Layout-aware segmentation

🔹 OCR & Document Processing

• Tesseract OCR
• pdf2image
• Poppler
• PyMuPDF (fitz)

🔹 AI & NLP

• Groq API (LLM reasoning)
• LangChain

🔹 Database

• Vector Indexing
• Full-text Search Index

---

⚙️ Key Engineering Components

---

1️⃣ PDF Processing Engine

Features:

• Converts PDFs to 300 DPI images
• Multi-threaded page conversion
• Handles scanned and handwritten PDFs
• Extracts both text and embedded images

---

2️⃣ Layout-Aware Segmentation (YOLOv8)

Instead of naive OCR:

• Detects text blocks using YOLO segmentation model
• Extracts structured regions
• Preserves logical content flow

This improves:

• OCR accuracy
• Content alignment
• Context grouping

---

3️⃣ Intelligent OCR Pipeline

Preprocessing:

• LAB color space conversion
• CLAHE contrast enhancement
• RGB normalization

Extraction:

• Tesseract OCR
• Page-wise concurrent processing

Improves:

• Low contrast scan handling
• Handwritten text clarity
• Reduced OCR noise

---

4️⃣ Parallel Page Processing

Uses:

• concurrent.futures
• Multi-threading for performance

Benefit:

• Faster evaluation for multi-page PDFs
• Scalable evaluation architecture

---

5️⃣ LLM-Based Semantic Evaluation

Instead of simple similarity scores:

• Sends extracted content to Groq LLM endpoint
• Uses contextual comparison prompts
• Produces structured evaluation output

This enables:

• Context-aware grading
• Semantic similarity detection
• Answer quality assessment