03 / 09
QA Agent
Pre-flight checks for high-volume email, automated.
- Role
- Architect
- Context
- Emirates, via Merkle, 2026
- Stack
- Node.js, Playwright, Vertex AI, Cloud Run, Cloud Build, Docker, Salesforce Marketing Cloud
QA Agent — narrated walkthrough
- 5–6d → 1–2dper QA cycle
- 100–200emails tested a day
- 93validation rules
The problem
Manual QA on high-volume, multi-market sends caught errors too late and consumed enormous effort. Rendering bugs, broken dynamic content and data mismatches slipped through.
What I did
- 1Built an agent that runs pre-deployment validation across links, dynamic content, rendering and data integrity.
- 2Automated rendering and behaviour checks with a Playwright-driven engine wired to SFMC.
- 3Surfaced results through dashboards, severity scoring and a version checklist for sign-off.
- 4Deployed inside Emirates’ Google Cloud: containers on Cloud Run, LLM checks through Vertex AI, run history in Cloud Storage, and CI/CD with Cloud Build.
Outcome
- A QA week that took 5–6 days of manual checking now takes 1–2.
- 100–200 emails tested within a day, against 93 rules — and a person still approves every test send.
- Every run is kept: fixes are re-tested automatically and the agent learns only from what the team approves.