Project

Multi Agentic Framework for Code Bug Summarisation and Evaluation

Objective

The increasing complexity of modern software systems makes error detection and correction challenging for developers, necessitating accurate and insightful error explanations. Generative Large Language Models (LLMs) offer automated solutions but often suffer from hallucination, lack of relevance, and repetitiveness, making manual evaluation labour-intensive and non-scalable. To address these challenges, we introduce a multi-agentic framework designed to automate the generation and evaluation of erroneous code explanations, ensuring clarity, accuracy, and domain-specific relevance. Existing solutions to evaluate machine-generated summaries fall short in domain specificity, failing to capture the semantic precision needed for accurate error identification and resolution in code

Outcome

Paper