Facial Recognition Attendance System Design

I want to put in place a complete attendance-management solution where every registered employee clocks in and out through a facial-recognition scanner. The moment the face is matched, the system must tag the exact time and GPS location.

Key workflow
• A valid record = matched face + in-time + out-time.
• If an out-time is missing after a configurable window, the entry is automatically flagged as invalid.
• Approvers need a dashboard that shows, per employee and per day, the in-time, out-time and total hours on site.
• Employees can send correction requests, and the system should decide automatically when the fix is obvious (for example, swapping two consecutive timestamps) while routing everything else to the supervisor for manual approval.
• Optional but desirable: capture a fresh live photo each day to further deter proxy attendance.

What I expect from you
1. A concise solution document describing the architecture, tech stack, database schema and user flows.
2. An analysis of scalability limits (more devices, larger face library, edge cases such as offline mode).
3. A concrete strategy for eliminating proxy attendance that goes beyond face matching—e.g., liveness detection or cross-checking device location.
4. Rough wireframes for the main user screens (employee kiosk, approver dashboard, correction module).
5. A working proof-of-concept that demonstrates the core flow: facial scan → in/out record → dashboard display. Non-core items may live only in the wireframes.
6. A short list of explicit assumptions made while building the above.

The deliverables should be easy to run locally (Docker, Vagrant, or straightforward setup) and include all source code plus deployment notes. Tech choice is open; feel free to lean on OpenCV, AWS Rekognition, Firebase, or any stack you are comfortable with—just justify the decision in the solution document.