Plate-Detection Model for Store Pickup: Training, Deployment, Store vs. Cloud Hosting

A retailer wants cameras at a store's pick-up area to detect the license plates of customers arriving to collect their orders. Explain how the model behind this would be trained and deployed, and how you would choose where it runs. ### Clarifying Questions - Does the model only need to locate plates in the frame, or also read them and match them to orders? - How would a customer's plate be known: entered at checkout, saved in the app, or learned on an earlier visit? - How many stores is this for, and do they already have cameras and dependable internet? - What privacy or retention rules apply to images of vehicles and plates where these stores operate? ### Part 1 — Training and deploying the model Explain how you would build the model, from data to a system running at the store. ```hint Start at the curb Picture the actual camera view at a pick-up bay across a full day and a full year, and work out what labeled data would have to cover it. ``` #### What This Part Should Cover - The problem framing and a success measure tied to the pick-up experience - Data collection and labeling that reflect real store conditions - The model approach, evaluation, rollout and monitoring ### Part 2 — On-premise or cloud How would you decide whether the model runs on equipment at the store or in the cloud? ```hint Trace one arrival Follow a single car from the moment it enters the frame to the moment staff are alerted, and note where a network round trip, an outage or a video upload would matter. ``` #### What This Part Should Cover - The criteria that decide where it runs, weighed for this use case - How the answer changes with the number of stores and their existing setup - A recommendation and a fallback for when the chosen path fails ### What a Strong Answer Covers - The customer and staff experience the model serves, and which kind of error costs more - Evaluation that goes beyond model accuracy to the end-to-end pick-up outcome - Privacy, consent and retention of plate data - What happens when the system cannot recognize an arrival ### Follow-up Questions - A customer arrives in a different car from the one on file. How should the product handle it? - Accuracy drops at one store after it changes its lighting. How would you notice, and what would you do? - How would you decide whether to build this model or buy an existing plate-recognition product?