Most product managers hand the data off. I’d rather read the model myself.

Payroll admins on Andal's enterprise HR platform needed to see which salary components were assigned to a specific employee.

The only path was looping through up to 100 transactions by reference number — 20+ employees each — so a single lookup could take ~22 minutes, and restart from scratch if the component turned out to be inactive.

During the desktop-to-web migration I preserved the familiar flow over my redesigned row-based view — familiarity over friction — and, when full cross-editing proved infeasible, settled on read-only visibility between single and bulk transactions.

A lookup that took up to 22 minutes of manual searching now takes seconds.

HR admins juggle dozens of employee-lifecycle deadlines — probation, contracts, BPJS, tax status — with no central tracker beyond sticky notes and spreadsheets.

A missed probation-to-permanent update had no reminder and no rehire path, so the only fix was a manual API call to the database — an oversight becoming a technical emergency.

Dropped a daily attendance reminder that would overload large-client systems, replacing it with a single trigger near the end of the payroll period — preserving the goal without the notification volume.

Consolidated 18 task types into one view and cut acting on a critical task from 6 steps to 2 — a 67% reduction.

A personal mood-tracking web app I'm building solo, from planning docs through to production.

Mood trackers either bury you in features or feel clinical. I wanted the smallest thing that helps you notice a pattern.

Deferred AI-generated reflections to a later phase and shipped a static, mood-keyed message set first, so the core loop works before adding cost and complexity. Routed sensitive keys through serverless functions instead of exposing them in the browser.

A working single-user MVP, architected so multi-user is a phase-two switch, not a rewrite.