case study · platform
Deploy Console
An internal deployment dashboard that replaced four shell scripts and a spreadsheet, and turned a ten-minute rollback into one click.
- Role
- Sole engineer
- Team
- Platform, 6 people
- Year
- 2026
- Stack
- TypeScript · Next.js · Go
the problem
A hundred engineers deployed through four scripts nobody owned, and the record of what shipped lived in a spreadsheet one person updated. When a release broke, finding the last good build took longer than fixing the bug.
what I built
One console over the existing pipeline: a live timeline of every service, the diff that went out, and a rollback that reverts to a named build. A Go service watches the cluster; the frontend reads a single event stream.
Decisions
Wrap the pipeline, don't replace it
Rewriting CI would have taken a quarter and blocked every team. The console reads and triggers the pipeline that already existed, so it shipped in six weeks and nobody had to migrate.
One event stream, not polling
Deploy state arrives over a single server-sent event stream. The UI holds no derived state, which killed a class of “dashboard says green, cluster says red” bugs.
Rollback as a first-class button
The riskiest action got the clearest affordance: pick a named build, confirm once, done. Before this it was a script argument people looked up in Slack.
Boring, readable config
Service definitions moved to a flat YAML file with schema validation, so a new service is a pull request a reviewer can actually read.