Own the document-intelligence platform that turns dense construction specifications into structured, retrievable knowledge. Migrated the AI stack to Google Vertex AI and built a custom LLM service on top of it. Designed a defense-in-depth LLM extraction pipeline for public-procurement scoring grids, with layered QA validation. Shipped a zero-shot document classifier to production with no labeled training data, integrated an OCR system that cut processing time by 90%, and delivered an end-to-end PDF-to-Excel pipeline for construction pricing documents. Introduced MLflow and Temporal to the team, making AI evaluation reproducible and observable.
Built AI-core infrastructure on the Runtime team to accelerate training, fine-tuning, and deployment. Architected and implemented a Go-based KServe integration for scalable, serverless inference, prototyped a Rust-based observability platform, and wired several MLOps tools into backend services to automate distributed training.
Established and led the MLOps practice on Kubeflow — CI/CD, production deployment, monitoring, and full model-lifecycle management — and stood up serverless inference with KServe fed by Kafka streams. Developed cybersecurity systems for real-time threat detection in cloud-native applications, combining time-series forecasting with a two-stage PyTorch + XGBoost pipeline on tabular telemetry. Built network-automation tooling with GNNs, LLMs, and Knowledge Graphs, and architected a multi-agent LLM system for automated network reasoning. Supervised a PhD intern, co-authored publications, and co-invented a patent (pending).
Designed and implemented an MLOps benchmarking tool — built on experiment tracking and pipeline orchestration — to track metrics and evaluate model quality across internet security, brand safety, and contextual ad targeting. Extended it for the product team to cover custom decision-rule evaluation against the IAB taxonomy and GARM standards. Also helped scale the inference system to process millions of videos and texts (YouTube, Twitter) in real time.
Engineered and deployed end-to-end ML solutions — fraud detection for OneSpan and Deutsche Bank, a computer-vision system for railway-tie defect detection (Roav7), multi-sport outcome forecasting, and NLP for resume-to-job matching. Sharpened real-time anomaly detection through temporal and behavioral pattern analysis, turned the sports-forecasting model ROI positive (~3% return), and owned the full pipeline from preprocessing and feature engineering through training, evaluation, and deployment.
Contributed to applied research projects: shallow-gas detection from seabed seismic imagery (BGS); multi-sensor intelligent machining process monitoring (AMRC, co-facilitator); semantic and instance segmentation of 3D point clouds (SenSat); and analysis of network metrics on customer perception of reliability (Telus).
Doctoral research in multi-messenger astrophysics, studying neutrinos and gamma rays with large-scale experiments such as ANTARES, HAWC, and KM3NeT. Focused on statistical analysis, detector performance optimization, and advanced methodologies for high-energy particle detection.
Researched gamma-ray astrophysics with H.E.S.S., covering detector calibration, performance studies, and data analysis.
Leading a non-profit genetic-genealogy platform dedicated to helping ethnic Circassians reconnect with lost ancestral ties and strengthen unity within the global diaspora. Contributed by researching over 1,000 surnames, building the analytical web portal, implementing AI chatbot interfaces, and developing real-time data infrastructure and dashboards.