Business Problem & Solution
ACSI · Hospitality · 2024–2025 · Public
The problem
On a busy restaurant floor the difference between a good experience and a bad one is often timing: an empty glass that sits, finished plates that linger, a table that is ready to turn but no one notices. Staff are attentive but stretched across sections, and the signals are easy to miss at peak.
The production system addresses this with overhead cameras that watch tableware (not people) and quietly nudge the right server at the right moment — refill, clear, offer another round, reset the table — so service feels effortless without adding headcount.
The solution (downscaled)
This demo is a faithful, downscaled reimplementation on synthetic, top-down, tableware-only frames — no people, privacy by framing. One vision stage reads each frame; everything else (tracking, aggregation, confirmation, rules, governance, dispatch) is deterministic.
The honest cost story is built in: only the vision stage is metered, a change gate skips unchanged frames, and a confirmation window kills single-frame false alarms before any alert reaches a human.
What it is not
No real cameras, no real staff notifications, no real venue. Every frame is synthetic and labelled; the employer is named only as context. The real detector, tracker and integrations are described in Architecture → Out of scope.