Ascend AI Judge: An AI-Powered Hackathon Evaluation Platform Using Large Language Models

Abstract

Hackathon events generate large volumes of student submissions that require timely, consistent, and expert-level evaluation. Manual judging processes are often resource-intensive, subjective, and difficult to scale. This paper presents Ascend AI Judge, a web-based intelligent evaluation platform that leverages Large Language Models (LLMs) to automate the assessment of hackathon projects across three key dimensions: abstract/project documentation, source code quality, and video pitch presentations. The system integrates a FastAPI backend with the Groq-hosted Llama 3.3 70B model to perform multi-modal analysis, producing structured scores, detailed feedback, and actionable improvement tips for each submission. A React-based frontend provides an intuitive interface for students to upload their work and receive instant, context-aware evaluations. The platform also features ARIA, an AI-powered mentor chatbot that engages students in personalized improvement dialogues grounded in their specific evaluation results. Experimental results demonstrate that the system generates consistent, rubric-aligned assessments with low latency, showing strong potential as a scalable supplement to human judging in academic and competitive settings.