Problem
The operational pressure, actors involved, and what breaks when state is unclear or ownership is missing.
← All work
Applied AI prototype
Edge-AI Seat State Monitor
Uncertainty handled before it becomes a decision.
Signals→
Classifier→
Guard→
Mode
PythonRandomForestSynthetic DataGuard Logic
System shape
Actors, actions, state transitions, boundaries, and the minimum structure required for the workflow to stay legible.
Technical decisions
Trade-offs across stack, data model, deployment, and guardrails — including what was intentionally deferred.
Implementation
How the system was built in practice: modules, interfaces, tests, and the path from prototype to usable product.
Limitations
Known failure modes, assumptions, and constraints that still shape how the system should be evaluated.
Next iteration
What would change under more time, scale, or production pressure — and why that order matters.
System layers
Feature inputs
pressure / loadseat accelerationaircraft / cabin refsensor validity
ML classifier
EMPTYOCCUPIED_STABLEOCCUPIED_MOVINGVIBRATION_EVENTSENSOR_FAULT
Guard logic
confidence threshold (0.65)pressure occupancy checksensor-fault routing
System modes
power savingcomfort monitoringdiagnostic monitoringfallback / no-action
System judgement
Low confidence does not trigger action.
It routes to fallback or no-action mode. The classifier informs — guard logic decides what the runtime may emit.