Starting Point

• A small platform team with aggressive deadlines.
• Prior attempts relied on nightly batch jobs that failed under load.
• They brought us in to design a real-time architecture that could ship quickly.

What We Built

• Event ingestion pipeline with schema validation at ingress.
• Real-time aggregation service and query API.
• Explicit in-scope: ingestion, storage, dashboards API. Out-of-scope: BI dashboard UI.

How AI Changed the Outcome

• AI-generated 340 test cases from the OpenAPI spec. Engineers accepted 280 (82%).
• Copilot-assisted CRUD generation saved roughly 60 engineering hours.
• Automated PR review flagged 23 security issues before human review; 19 were true positives.

What We’d Do Differently

We over-invested in AI-generated integration tests in week 2. A mutation-testing-first strategy for unit tests would have surfaced the same defects with faster feedback.