Lead ML Scientist/Engineer
Jul 2022-PresentAchieved a 50% increase in our ML model's ability to detect weapons carried above 5ft by incorporating object localization as a prior.
Conceptualized and executed "the winning framework" to define metrics based on complex customer requirements, and translate those requirements into clear, quantifiable questions that our team could answer through data, testing and modeling.
Orchestrated the integration of advanced ML models onto NVIDIA edge computing devices, by converting data cleaning processes, feature engineering, deep residual networks and xgboost models to ONNX pipelines, subsequently deploying models onto NVIDIA Triton Inference Server.
Drove a 5% revenue increase and reduced hardware downtime by 14 days on average by developing a real-time analytics suite to diagnose hardware failures. Analytics were developed in Python using PCA, semidefinite programming, and frequency-based signal processing.
Implemented a cutting-edge dashboard for diagnostic monitoring, leveraging Docker and Prometheus technologies, that streamlined engineer operations and maximized system uptime.
ML Scientist/Engineer
Jan 2021-Jul 2022Designed and deployed a robust Sqlalchemy/PostgreSQL data model within a Dockerized environment, improving data selectivity at train and evaluation time, which led to an increase in training data quality and an 8% reduction in false positives.
Trained, validated and deployed residual networks that were able to reduce false alerts due to steel toed boots at factories by 30%, while crucially maintaining weapons detection capabilities.
Increased the ROC AUC of threat detection xgboost models running in high interference environments by augmenting training data with on-site noise and using correlated sensor information to remove far field interference.
Refined object localization algorithms, reducing MSE by 20%, which streamlined the customer screening process, directly impacting operational productivity.
Data Scientist
May 2019-Jan 2021Increased the mean average precision of our image recognition deep networks by 12% through refinement of image data and development of image augmentation software in Python.
Using PyTorch ResNets for person detection and random forest regressors, led the end-to-end development of a social distancing detector accurate to within 0.1m from ideation to deployment. Led to US patent (US20220172483A1).
EDUCATION
University of Toronto
Master’s Degree: Electrical and Computer Engineering
(Expected May 2024)
Coursework: Biomedical Signal Processing, Statistical Methods for Machine Learning, Deep Learning Theory, Convex Optimization, Detection and Estimation Theory, Bio-inspired algorithms for Smart Mobility.
Research: Investigating the potential of ECG-derived breathing signals as predictors for SUDEP, aiming to set a new precedent in patient monitoring and preventive healthcare.
University of New Brunswick
Bachelor’s Degree: Computer Science (Graduated 2021)
Summa Cum Laude
Minor in Mathematics
Minor in Statistics
Honor's project: Streamlined occupational coding for healthcare professionals through the development of a high-accuracy, high- throughput NLP pipeline utilizing TF-IDF and Doc2Vec embeddings. Research presented at 28th International Symposium on Epidemiology in Occupational Health (Abstract).