Architected and shipped the full video AI pipeline for an embedded product — from GStreamer pipeline design through custom model quantization down to on-device inference on the Qualcomm Neural Processing SDK.

Sound-based ML system to detect my doorbell - covering the full lifecycle: data collection & management, model training, and self-hosted deployment. A progressing project, not a notebook demo.

End-to-end automated training and evaluation pipeline for a driver-assistance function. Includes a custom time-series data management framework for car trip data, integrated with Azure ML and internal MLOps tooling.

Clean Python implementation of the MUSIC algorithm (MUltiple SIgnal Classification) for direction-of-arrival estimation. Shows the algorithmic depth beneath the CV/ML work — rooted in classical signal processing theory.