I developed edge-AI at Good Systems, a research initiative focused on empowering skilled trade workers through human-centered machine intelligence.

As an intern, I co-designed and compressed a TensorFlow Lite model (~400 KB) to run on a low-power microcontroller. The model achieved an ~83% inference accuracy for classifying Dremel usage modes (e.g. sanding, rastering, cutting, or drilling), initiating our implementation of real-time feedback. During development, I also implemented a structured MLOps workflow using GitLab, TensorBoard, and MLflow to systematically version models and monitor performance across training iterations. Training data was collected through standard use of construction tools.

To bridge backend analytics with user experience, I engineered a Bluetooth Low Energy (BLE) communication and paired it with a custom iOS app that visualizes tool orientation and activity in real time, making smart insights accessible on handheld devices.