Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{198172,
        author = {Binish Moosa and Nishad Joglekar and Samar Khan and Kashif Khan and Prof. Shiburaj Pappu},
        title = {Multi Agent AI-Powered Codebase Debugger & Analysis System},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {13525-13529},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=198172},
        abstract = {Comprehending large-scale multilingual codebases presents a formidable bottleneck in the software development process, accounting for more than 50% of a developer’s time. Current static analysis frameworks are limited in their semantic capabilities, while cloud-based large language models have serious privacy risks. This paper presents NeuroSense, an offline, multi-agent framework that combines deterministic static analysis and a locally hosted language model (Llama 3.1:8b) to provide human-comprehensible reports on codebase architecture, execution flows, and potential security threats. The ingestion pipeline leverages regex-based parsing techniques, supplemented by Python Abstract Syntax Tree (AST) for optimal precision, to analyze modules, call graphs, dependency graphs, and symbol tables from over 20 languages. The three distinct agents - Architecture, Logic (Developer), and Security - communicate through LangGraph to deliver parallel, structured reports. A key-facts system is used to minimize LLM hallucination at 1.8%, and fallback templates allow for graceful degradation. Tested on real-life projects (Flask, Express.js, Tokio), NeuroSense provides a call resolution rate of 96.4% for Python, 87.5% overall, and an F1-score of 84.2% for vulnerability detection. An end-to-end analysis takes approximately 55 seconds for medium-sized codebases. The proposed system works offline, thus guaranteeing privacy and improving developer onboarding by 60-70%.},
        keywords = {Code analysis, multi-agent system, large language models, static analysis, software security, offline LLM.},
        month = {April},
        }