Face Recognition Attendance (Kiosk + Mobile)

ECOSIRE builds a proper Frappe app (installed with bench get-app and bench install-app, packaged for Frappe/ERPNext v15 and v16) that adds that missing identity layer without forking core HR. We add custom DocTypes — a Face Enrollment record holding per-employee face templates as embeddings (never raw photos in plain storage), a Kiosk Device registry, and a Face Checkin Log — plus whitelisted server methods (@frappe.whitelist()) exposed over the Frappe REST API that a tablet kiosk PWA or a mobile selfie screen calls. On capture, the client posts a frame; a liveness / anti-spoofing pass rejects printed photos and screen replays; the embedding is matched against enrolled templates at a configurable 95–100% confidence threshold; on a match the method creates the Employee Checkin exactly as ERPNext expects, so Shift Type assignment, auto-attendance, and payroll continue to work downstream with zero changes to your existing flow.

Technically, enrollment and matching are wired through hooks.py doc events and scheduler events: a doc_events hook on Employee can trigger an enrollment reminder, and a scheduler_events job reconciles offline captures and prunes stale device tokens. Client Scripts drive the kiosk and selfie UI validation, Server Scripts (where you prefer low-code) can enforce geo-fencing or shift-window rules, and everything respects Frappe permissions — we ship dedicated Role Profiles (Kiosk Operator, HR Manager) so an operator device can create check-ins but never read salary data. Offline capture is a first-class feature: the kiosk queues punches locally when the network drops and syncs them through an idempotent whitelisted endpoint once connectivity returns, so a warehouse dead-zone never costs you an attendance record.

Because this is build-to-order, we start from your real constraints: number of sites, whether you enroll from HR-supplied ID photos or live capture, your matching provider (on-device embeddings vs. a self-hosted inference service you control for data residency), and how tightly you want geo/shift rules enforced. ECOSIRE scopes it, builds it on a staging bench, runs UAT with your HR team, then installs it on your production ERPNext and hands over the git repository. Typical delivery is 2–4 weeks from confirmed scope — there is no instant download; this is a bespoke app we build, install, and support for you.