Case Study
DevScope
An engineering analytics platform that tracks PR velocity, review lag, and code churn across GitHub repositories. Streams live events through a GCP pipeline into BigQuery and automatically flags productivity anomalies using Vertex AI.
6-service event-driven GCP pipeline
Rebuilt anomaly detection with per-repo rolling baselines
Sub-second analytics on live streaming repo events
What
Built a distributed engineering analytics platform mining GitHub repository data to surface PR velocity, review latency, and code churn.
Streams events through Cloud Pub/Sub into BigQuery for sub-second querying.
Vertex AI anomaly detection on Cloud Run flags productivity regressions across tracked repositories.
Why
PRs sat for days. Reviews piled up. Nobody knew where delivery slowed down across the repositories they owned.
Who
Engineering managers tracking delivery health, repo owners identifying bottlenecks, and platform teams monitoring engineering velocity.
When / Where
Most useful for live repository monitoring, high-volume webhook ingestion, and teams that need fast operational dashboards.
Constraints
GitHub webhook traffic is bursty, so the ingestion layer had to handle spikes without dropping events.
Dashboard queries needed to feel fast, not batch-delayed.
Wanted to avoid managing servers while still having real infrastructure flexibility.
What
Built a distributed engineering analytics platform mining GitHub repository data to surface PR velocity, review latency, and code churn.
Streams events through Cloud Pub/Sub into BigQuery for sub-second querying.
Vertex AI anomaly detection on Cloud Run flags productivity regressions across tracked repositories.
Why
PRs sat for days. Reviews piled up. Nobody knew where delivery slowed down across the repositories they owned.
Who
Engineering managers tracking delivery health, repo owners identifying bottlenecks, and platform teams monitoring engineering velocity.
When / Where
Most useful for live repository monitoring, high-volume webhook ingestion, and teams that need fast operational dashboards.
Constraints
GitHub webhook traffic is bursty, so the ingestion layer had to handle spikes without dropping events.
Dashboard queries needed to feel fast, not batch-delayed.
Wanted to avoid managing servers while still having real infrastructure flexibility.