Machine Learning Software Engineer at Motive Technologies Inc.

Built Variational Autoencoder (VAE) foundation model from scratch on internal dashcam video data, enabling scalable, unsupervised scene understanding in future perception models. Designed downstream scene classification tasks to benchmark learned embeddings; analyzed latent space via t-SNE, PCA, & clustering to surface semantic structure. Adapted MobileNetv2 - UNet architecture to minimize model size, targeting feature-rich edge deployment. Launched internal research initiative & literature review on scene graph learning; synthesized key insights that drove strategic pivot from traditional object detection to contextual & relational scene modeling.

Machine Learning Software Engineer at NuraLogix Corporation

Sep 2022 – Dec 2022: Continued development on video-based vitals estimation, exploring signal processing and robust modeling. Implemented initial versions of a bio-signal pipeline and explored priors for noise robustness; evaluated denoising approaches to improve signal quality in challenging conditions.

Machine Learning Software Engineer at NuraLogix Corporation

May 2023 – Aug 2023: Engineered end-to-end signal extraction pipeline to estimate heart rate, breathing rate, & other vitals from driver videos, leveraging Fourier transforms, bandpass filters, & blind source separation methods. Boosted bio-signal robustness under noise by adapting Gaussian Mixture Models & transferring priors from audio source separation models to the video domain. Finetuned & benchmarked PulseGAN to denoise & enhance weak bio-signals in low-light & other edge conditions.

Data Scientist at Woodbridge Foam Corporation

Processed & cleaned 100+ GB of raw proprietary test data into structured formats for manufacturing optimization. Prototyped predictive time series models to reduce Indentation Force Deflection test time by over 90% – laying groundwork for faster QA cycles & substantial future cost savings. Used AutoML ( FEDOT ) for time series model selection, reducing manual tuning time & accelerating experiments.

Robotics Software Engineer at Aquadrone (University of Waterloo) – Student Design Team

Architected robot's vision pipeline system from ground up using OpenCV & ROS 2 for underwater object detection. Created data labelling & augmentation tools to fine-tune off-the-shelf YOLO model on mission environment data.