Case studies · AI workflow · peoplepuls.com
Managers stop writing reviews from memory
People Pulse drafts the review from a full cycle of GitLab, Jira, and Slack metadata — never message bodies. Managers edit. They never start blank. Four-week pilot on their own tools.
What changes
Evidence assembles across the cycle: merged MRs, incidents, onboarding, calendar. The manager gets a draft they can defend. Calibration and retention sit on the same trail. Pilot is four weeks on one team’s actual tools.
What I built toward
- AI-drafted reviews from work signals (GitLab, Jira, Slack, calendar).
- Privacy constraint as architecture: metadata only — never message or email bodies.
- Multi-company isolation, SSO, audit trail — so the workflow could be sold, not just demoed.
- Keep the review/calibration workflow; replace the foundation underneath (files-on-disk → durable store) without a rewrite.
Who it’s for
People/HR at a 150–800 person tech company that already lives in GitLab and Slack, and is tired of Lattice-priced tools that still start from a blank review.
Live product: peoplepuls.com.