PreVise: Multi-Agent Code Review

Modern software teams rely heavily on Pull Requests for code review, but manual reviews can be time-consuming and may overlook security vulnerabilities, logic errors, inadequate testing, and documentation gaps. Developers often need multiple tools to identify these issues, while existing automated checks may focus on only one aspect of code quality. PreVise addresses this challenge through an AI-powered, multi-agent Pull Request review system. It uses IBM Bob as the orchestration layer to coordinate specialized agents for four key areas: Sentinel for security vulnerabilities, Logic Agent for programming and business-logic errors, TestPilot for testing gaps, and DocsGuard for documentation inconsistencies. The target users are software developers, development teams, and project maintainers who want faster and more comprehensive PR reviews. Users configure the repository, Pull Request, branch, author, and local path through the PreVise dashboard and initiate a review. The system analyzes the code, aggregates findings, and presents each issue with its severity, location, evidence, and suggested remediation. A key feature is the human-in-the-loop fixing mechanism. Developers can approve or reject proposed fixes rather than allowing changes to be applied blindly. Approved fixes are applied by the Fix Agent, followed by an automated re-scan and regression-test gate to verify that the issue has actually been resolved. What makes PreVise distinctive is the combination of multi-agent analysis, explainable findings, human approval, automated remediation, and post-fix verification within one workflow. Instead of simply reporting problems, it creates a closed-loop review process from detection to verified resolution. By parallelizing specialized checks and presenting their results through a unified dashboard, PreVise reduces repetitive manual review effort while helping teams identify critical issues earlier and make more informed decisions before code is merged or deployed.