I need an end-to-end face-recognition attendance solution that splits its workload between a Raspberry Pi at the edge and a lightweight cloud frontend. On the edge – The Raspberry Pi must run an optimised computer-vision model that can detect and recognise multiple faces in the same frame, then immediately tag each recognised person with a timestamp. All inference happens locally; only the processed results travel to the cloud, never the raw video. Model size, frame rate and power draw have to suit a typical Pi 4 with a standard camera module. Connectivity & data flow The Pi should expose a small, secure REST or MQTT service that pushes attendance events to the cloud and retries automatically if connectivity drops. A simple enrolment script should let me add new faces directly on the Pi or via an API call. Cloud side The frontend will live on my preferred cloud stack (a basic React or Vue single-page app is fine) and must focus on a clean Attendance Records view: searchable, filterable by date/person and exportable to CSV. Authentication can be as simple as email/password for now. Future-proofing While the immediate requirement is multiple face recognition, please structure the code so that real-time dashboards or anti-spoofing modules could be slotted in later without major refactoring. Deliverables • Trained face-recognition model and Python inference script for Raspberry Pi • Edge service for data push with brief deployment guide • Cloud API (Node, Python or Go) plus single-page UI showing Attendance Records • Read-me and comments that let a moderately technical user redeploy everything from scratch Acceptance criteria 1. System recognises at least five faces simultaneously with ≥95 % accuracy in indoor lighting. 2. Attendance event appears in the cloud UI within five seconds of capture. 3. Full installation from a clean Pi OS image to a working cloud dashboard can be completed using the supplied documentation alone.