AI-driven observability platform for application reliability

Abstract

In recent years, AI-assisted software development has been widely adopted, increas￾ing code production and accelerating delivery cycles. While this improves developer productivity, it also introduces new challenges in code review, quality assurance, and runtime reliability. Existing solutions typically address isolated stages of the software lifecycle, lacking a continuous connection between pre-merge analysis and post-deployment incident response. This fragmentation motivates the need for a unified platform that integrates code review and production observability within a single workflow. This thesis presents Falcon, an AI-driven observability platform designed to bridge this gap by unifying pre-merge and post-deployment processes. The system integrates a codebase graph engine for static dependency analysis using ts-morph, a sandboxed execution environment for secure code exploration, GitHub integration, and agent-based reasoning for structured code review and automated fix generation. It enables cross-file defect detection during pull request analysis, root- cause investigation of runtime failures through Sentry integration, and automated patch generation with type-checking validation. A case study conducted on a pro duction TypeScript monorepo demonstrates the feasibility of unifying code review and incident remediation within a single AI-assisted workflow. By combining repos itory analysis, runtime observability, and agent-based reasoning, Falcon supports continuous software quality assurance and contributes to improving the reliability and maintainability of modern software systems.