Case 15 · CI/CD LLM Code Review + Unit-Test Generation · Code Review P50 24h → 1h · Double-Digit Defect-Escape Reduction¶
TL;DR: CI/CD LLM Code Review + Unit-Test Generation · Code Review P50 24h → 1h · Double-Digit Defect-Escape Reduction. Direction: AI + DevOps. Scale: 200+ engineers, 50k commits/month. Hard KPI: Code Review P50 24h → 1h, double-digit defect escape reduction. Tech: GitLab webhook, LLM router (DeepSeek + GPT-4o), unit test auto-generation. My role: platform architecture, CI/CD plugin, model routing, effect acceptance. Client name anonymized under NDA.
- Direction: AI + DevOps
- Industry: SaaS
- Scale: 200+ engineers, 50k commits/month
- Hard KPI: Code Review P50 24h → 1h, double-digit defect escape reduction
Problem¶
200+ engineers, 50k commits/month, Code Review queued P50 24h, high-risk changes leaking through.
Solution¶
GitLab webhook triggers, LLM static analysis + recent-30 commit correlation + rule-library match, high-risk to human, mid/low auto-approve.
Tech Stack¶
GitLab webhook, LLM router (DeepSeek + GPT-4o), unit test auto-generation
Result (Hard KPIs)¶
Code Review P50 24h → 1h, double-digit defect escape reduction
My Role¶
platform architecture, CI/CD plugin, model routing, effect acceptance
Related¶
Next: Download PDF · 30-min alignment
Client name anonymized · KPIs verified · Real client name disclosable under NDA.