Clocking in a crew, even with no signal.

Short intro screens say the one thing a supervisor needs to hear: clock your crew in, even offline. Then phone and PIN sign-in, an SMS code, and a permissions screen that explains why each permission is needed.

The language switcher is on the very first screen. Someone who can't read the default language won't find it in settings.

Home opens on the current site and shift, clocked-in against expected, and four big actions. A sync banner shows exactly how many records are waiting to upload.

Switching site is a bottom sheet, because supervisors often cover two or three.

Clocking in is multi-select: tick the crew, confirm the location, clock everyone in at once. With no signal, the app says “saved on your phone” with a count and an illustration, never a red error.

When HR declines a record on web, the reason lands here. The supervisor corrects the time and resends from the same screen.

Three steps: type and placement, then a national ID scan with a live NID match, then phone. Offline, the worker is saved locally and marked “pending sync”.

The success screen offers “register another”, because hires come in batches.

Approve or decline leave from a list, request leave on someone's behalf with a photo of a doctor's note, and adjust per-area productivity rates in a bottom sheet.

I designed every state the scan can be in before designing any flow. Copy is short and plain, because it's read at arm's length while holding a phone up to someone's face.

A confident match, saved and ready for the next person. Clock-out. No match, with manual search. Output recorded against a face-matched worker. And finding an employee by face.

Enrolment is step five of registering a worker. It starts with consent, captures the face, then checks for duplicates. A close score sends it to a supervisor; a very high one blocks it.

A timed-out duplicate check never silently passes. The worker stays unenrolled until the check completes.

The best decisions here came from watching supervisors work rather than from the brief. A sync count on the home screen did more for trust than any explanation of how offline works.

With face recognition, the hardest call was to make the app less confident. Designing the uncertain states first (lookalike, no match, timed out) is what made the confident ones safe.

Next

• Ship biometrics Phase 1 and track wrong-worker records and scan time weekly against the targets.
• Field sessions in Kinyarwanda focused on offline sync and the lookalike state.
• Resume onboarding for the paused workers once enrolment holds up in the pilot.

Single screens at full size, including the face enrolment card on a worker's profile.