Three sequential steps must all pass: GPS geo-fence check (within 300m) → Liveness challenge (randomised blink or head-turn, in-memory only) → Face match against 128-D Azure embedding (confidence ≥ 0.6). All three must pass for punch to be recorded. Failure in any step blocks the punch.

Before liveness or face match, GPS coordinates are validated against the site's configured geo-fence boundary (default 300m). If outside → punch immediately blocked, FCM P1 alert to Supervisor + Admin, anomaly logged as ANOMALY_GEO. All three anomaly types — Geo, Ghost, and Duplicate punch — shown here.

After geo-fence passes, a randomised liveness challenge (blink or head-turn) is presented. Azure Liveness API processes the live camera feed in real-time — entirely in-memory, nothing stored at this step. Randomisation prevents pre-recorded video replay. Failure shows a clear retry prompt. Both challenge types shown below.

Confidence ≥ 0.6 → punch recorded with GPS, timestamp, location photo (S3). Shift activated, Supervisor notified via FCM. Confidence < 0.6 → 2 retries allowed (3 total). After 3 fails → Supervisor must manually override. Multiple punches supported — late arrival, mid-shift break, early departure. All logged separately.

After punch-in, HK Staff sees their task list for the shift — assigned by Supervisor before or at shift start. FCM push notification delivered with task list. Each task shows location, instructions, SOP reference, and due time. Staff marks done in app — optional photo evidence. Incomplete tasks at shift end auto-flagged to Supervisor and Field Officer.

Staff marks tasks done in the app. Optional photo evidence captures the completed work area — supervisor sees real-time update. Photo captured in-app only, GPS + timestamp embedded. PATCH /task/{id}/complete syncs to supervisor immediately. Overdue tasks stay highlighted until resolved.

HK Staff don't conduct their own check — the Supervisor photographs them during floor rounds. However, the app shows the staff member their current compliance status and any violations. Safety reminders from SOPs are surfaced here — wet floor sign, PPE, chemical protocol — reinforcing what was taught in training.

The app uses the device's built-in gyroscope and accelerometer to passively monitor staff movement during the shift. Detects unusual patterns — prolonged idleness (possible sleeping), abnormal sudden movement (possible fall or accident), or suspicious stillness when tasks are assigned. Alerts are sent to Supervisor and Field Officer automatically. Visible passively to staff.

At shift end, HK Staff completes the same 3-step face recognition process for punch-out. GPS, liveness, and face match all re-verified. Punch-out time recorded with GPS coordinates and location photo. Total hours calculated. Supervisor notified via FCM. Attendance record finalised for payroll engine.

The HK Staff app handles no-network scenarios using SQLite local cache. Tasks are pre-loaded at shift start. Punch-in attempts use last known GPS (cached). Photos are queued for S3 upload. Task completions stored locally. Everything syncs automatically on reconnect. Offline sync marked in the attendance record.

Staff can view their own attendance records — punch-in times, punch-out times, late arrivals, absences, and GPS accuracy flags. data_scope: own_only means attempting to view another employee's records returns HTTP 403. Monthly summary shows working hours, absenteeism, and late count used for payroll calculation.

Monthly payslip generated on the last working day of the month. Shows gross salary, all statutory deductions (PF 12%, ESI 0.75%), and non-statutory deductions (absent days, late arrivals, advance, dress allotment if applicable). Net pay shown prominently. Staff can download PDF. FCM notification when payslip is generated.

HK Staff can view their own training history and current skill levels. Shows sessions attended, quiz scores, and which sessions are upcoming. Remedial flags appear here if quiz score was below 80%. Skill levels updated after each training session by the Trainer. Staff cannot modify their own skill levels — read-only view.

Digital ID card auto-generated after onboarding completion and admin approval. Contains staff name, employee ID, photo, site, designation, and QR code linking to the encrypted employee record. Staff can show this for identification at site. Document status (Aadhaar, PAN, police verification) visible here with expiry warnings.

HK Staff receives the most limited set of notifications — only things directly relevant to them. P3 events include task assignments, training session reminders, and document expiry warnings. P4 includes payslip generation. They do NOT receive P1 safety/security alerts, SLA breach alerts, or cross-site notifications.