CodeGrade: An AI-Augmented Automated Code Grading and Mentoring Platform Integrating Multi-Dimensional Evaluation, Socratic Tutoring, and Gamified Arena Mechanics
Year: 2026
Type: article
Abstract: Traditional automated code evaluation systems in Computer Science (CS) education predominantly rely on deterministic input-output string matching. While effective for assessing basic execution correctness, classical autograders fail to evaluate code readability, structural logic, algorithmic efficiency, and Big-O asymptotic complexity. Furthermore, students receive cryptic error tracebacks without pedagogical guidance, leading to reduced engagement and persistent anti-patterns. To overcome these constraints, this paper presents CodeGrade, an end-to-end AI-augmented code evaluation and mentoring platform. CodeGrade integrates a Web-based Microsoft Monaco VS Code editing engine, an isolated dual-sandbox execution environment (Judge0 containerized runner), and a Large Language Model (Groq Llama 3.3 70B) into a unified microservices architecture. The platform introduces a multi-dimensional 60/20/20 Grading Matrix that balances functional correctness (60%), AI structural logic quality (20%), and Big-O asymptotic complexity alignment (20%). Additionally, CodeGrade incorporates an AI-powered Socratic Logic Tutor enforcing strict zero-code-leakage constraints, a Monaco Line-by-Line Refactoring Diff Viewer, Web Speech API Text-to-Speech audio reviews, and a Gamified Peer-to-Peer (PvP) Battle Arena. Empirical evaluation demonstrates an average execution latency of 11 ms, high grading consistency (98.4%), and significant improvements in student engagement and code optimization quality.
Authors: G, Yeswanth, Vimalraj, Govindaraj, Thilak, R C
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7222561382
FWCI:
Cites:
Cited by:
Related to: