Edge AI · Edge and cloud computing for AI
Design the pipeline between the sensor and the decision.
Each scenario places the sources and sinks for you. You choose the edge devices, gateways, on-premises servers and cloud services between them, where each AI stage runs and in which model format, and how every link carries its data. The sandbox computes latency, bandwidth, monthly cost, device energy, accuracy and privacy as you build, and checks your design against the scenario's constraints.
Every number opens into the formula that produced it, with your values substituted, so you can check it by hand. Your lecturer grades the reasoning in your written justification, not the diagram.
Practice scenarios
Practice designs save in this browser. Nothing is sent to a server.
- S1Ward fall detectionHospital, Klang ValleyCeiling cameras watch for patients falling. The alarm must sound at the nurse station within a second, and video may not leave the hospital.LatencyPrivacyOfflineBudgetAccuracy
- S2Motor predictive maintenanceE&E plant, PenangVibration sensors on 200 motors feed an anomaly detector that can stop the production line before a bearing fails.LatencyBudget
- S3Oil palm fruit gradingCollection ramps, SabahCameras grade the ripeness of each fresh fruit bunch at 30 remote ramps with patchy 4G coverage.LatencyOfflineBudgetAccuracy
- S4Adaptive traffic junctionKuala LumpurCameras at four junctions count vehicles per lane so the signal controller can adjust green times in real time.LatencyPrivacyBudgetAccuracy
- S5Flood early warningRiver basin, KelantanBattery-powered river stations feed a basin-wide forecast that can sound sirens before the water arrives.LatencyBatteryBudget
- S6Federated retinal screeningPrivate clinics, nationwideFifteen clinics screen eye images for diabetic retinopathy and want to improve the model without pooling patient images.LatencyPrivacyTrainingBudget
Three tiers, by colour and by name
Edge on or beside the sensor, Fog on site, Cloud off the premises. Placement drives every metric.
An analytic model, not a simulator
Results come from closed-form formulas with an M/M/1 queueing approximation. They are teaching approximations: good for comparing designs, not for buying equipment.
For lecturers
Create classes, collect submissions and export marks from the instructor dashboard.