International Journal of Science, Engineering and TechnologyAuthors: Akshat Garg, Rahul Kumar, Abhishek Kumar, Pratiksha Singh
Abstract: Software quality assurance has traditionally depended on manual code review processes and heuristic-based static analysis tools, both of which suffer from scalability limitations, reviewer fatigue, and inconsistent coverage. This paper presents a comprehensive GenAI-Powered Full Stack Code Review and Bug Prediction System that integrates large language models (LLMs) with static analysis, abstract syntax tree (AST) parsing, and retrieval-augmented generation (RAG) to deliver automated, context-aware code reviews alongside probabilistic bug predictions. The system spans the complete software development lifecycle, embedding seamlessly into Git-based workflows via CI/CD pipeline hooks and IDE plugins. Our architecture employs fine-tuned Code-LLaMA and GPT-4o models for multi-language code understanding, a graph neural network (GNN) for inter-module dependency analysis, and a gradient-boosted classifier for historical defect pattern recognition. Experimental evaluation across five open-source repositories totalling over 2.3 million lines of code demonstrates bug detection precision of 87–93%, false positive rates below 8%, and an average review latency under 4 seconds per file. Developer surveys indicate a 44% reduction in post-merge defect density and a 38% improvement in review turnaround time. The system’s explainability layer, built on attention visualisation and chain-of-thought reasoning, ensures AI-generated feedback is transparent and actionable.
