MaskGuard

The problem

What the product needed to solve

Real-time safety checks need to operate quickly and consistently without sending every camera frame to a remote service.

Constraints

The realities shaping the work

• Run inference within the limits of a mobile device.
• Keep the camera and prediction loop responsive.
• Present uncertain machine-learning output responsibly.

Approach

Product and technical direction

A Flutter application that integrates a TensorFlow Lite model for local inference and turns predictions into an immediate, understandable interface.

Architecture

Clear boundaries between responsibilities

Camera input is prepared for a compact TensorFlow Lite model, inference stays on device, and the UI consumes normalized predictions rather than depending directly on model-specific output.

Implementation highlights

Where the engineering work mattered

• Integrated mobile camera input with an on-device inference workflow.
• Kept the machine-learning layer separate from interface state.
• Avoided a remote inference dependency for the core detection loop.

Outcome

What the work established

• Produced a working demonstration of practical on-device computer vision in Flutter.