Inspiration
Analytics teams ship SQL and dashboard changes daily, but review, data-quality testing and privacy checks are manual and easy to skip. One bad query or leaked column reaches leadership dashboards before anyone notices.
What it does
On every push to SQL/dashboard files, GitLab Duo Agent Platform flows:
- Review the query for logic and performance issues
- Triage data-quality test results
- Guard against leaked personal data (PII)
- Write plain-English release notes
- Post a post-pipeline summary
A human approves only the release (manual release_approval job) before deploy to GitLab Pages.
How we built it
- GitLab CI/CD: lint → test → security (SAST, secret detection) → build → deploy, with a manual approval gate
- GitLab Duo Agent Platform flows with MR-approved and pipeline triggers
- GitLab Pages for the live dashboard
- Demo scripts: inject_bad_change.sh / reset_demo.sh
Challenges
- No "MR merged" trigger, so flows run on MR approved and pipeline pass/fail
- Unlocking Duo Agent Platform under the deadline (Ultimate + Duo trial on a group)
Accomplishments that we're proud of
Five working GitLab Duo flows (query review, PII guard, data-quality triage, release notes, post-merge test summary) wired to real MR and pipeline events, a green CI pipeline, and a live dashboard on GitLab Pages — built end to end in under two days.
What we learned
How to design supervised agent flows with clear triggers, keep AI review explainable for data teams, and combine Duo agents with ordinary CI checks so humans stay in control.
What's next for DataGuard Flow
Support more databases (Postgres, BigQuery), custom PII rules per team, Slack/Teams alerts, and a one-click template so any GitLab project can add DataGuard checks.
Log in or sign up for Devpost to join the conversation.