An automated AI-powered code review gate on GitLab CI with confidence filtering and token budget guardrails.
Role
Creator & Lead Engineer
Year
2024
Stack
TypeScript · GitLab CI · LLM Guardrails · Docker
Team
Developer Platform
MergeScout is an automated code review intelligence gate running inside GitLab CI pipelines, delivering precision-calibrated code feedback, security defect detection, and actionable suggestions on merge requests.
Problem
Engineering teams at scale face review bottlenecks and noisy automated linting tools that generate false positives, leading to developer fatigue and wasted LLM token budgets on low-signal files or formatting churn.
Approach
Built an intelligent review orchestrator using TypeScript and GitLab API integrations. Implemented multi-stage semantic diff filtering, prompt token budgeting, AST-aware context slicing, and a precision/recall feedback loop evaluated against senior engineering review benchmarks.
Outcome
Reduced code review turnaround time by 35% across participating frontend and backend teams
Filtered out 85%+ of low-confidence AI noise, maintaining high developer trust
Kept LLM inference costs strictly bounded with deterministic token budgeting
