Anyone can make an app in a few minutes now with Claude. It's boring.
Instead we should be pushing AI to its boundary, so that it's faltering halfway between hallucination and true novelty...
Because what is innovation but a hallucination we set out to make real?
PhD researcher specializing in computer vision and robotics, deploying advanced vision technologies from research to production.
First-author publications in top-tier conferences (ICRA, RA-L) with patented technologies used in production systems worldwide.
Leveraging AI to build AI. Deploying production-grade AI products in real-time systems across multiple industries and countries.
Featured Projects
🔬 Research Innovations
- • Novel machine-readable fiducial markers under deformation
- • Real-time surface reconstruction and shading estimation
- • Custom Metal GPU compute kernels achieving > 60fps on iPad
- • Deployed in 8+ countries with enterprise clients
🔬 AI Innovations
- • Multi-agent orchestration for complex educational workflows
- • Advanced context management for large video processing
- • Specialized agents for content extraction and question generation
- • Personalized learning pathway optimization
🔬 Technical Innovations
- • Real-time 3D shape analysis of irregular organic objects
- • Optimal bin packing algorithms for maximum density
- • Deployed to agricultural facilities in Israel
- • 2-second cycle times with 99%+ reliability
🔬 Technical Innovations
- • Data-embedded markers that maintain readability under deformation & tears
- • Real-time surface tracking with sub-millimeter precision
- • Novel encoding schemes for robust data recovery
🔧 Core Technologies
- • quantized YOLOv11n & Custom pose estimator
- • Kalman filtering + Batch bundle adjustment
- • Domain-randomized training & Real-image self-supervision
- • <30ms latency on-device with standard iPads
Novel online self-supervision method targeting real runtime environments for 6D object pose estimation in industrial bin-picking. Achieves state-of-the-art performance with correct detection rates above 95%. Features multi-view ground truth estimation, synthetic hard-case training, and active target selection in the real domain.
Advanced research project exploring cooperative AI systems, real-time decision making, and natural language command processing in complex 3D environments. Features sophisticated behavior matrices and team coordination algorithms.
Advanced AI system featuring 7 distinct LangGraph workflows for context analysis, web research, and intelligent session management. Real-time web research integration with context preservation.
Features real-time 3D model generation of tools and equipment with unique segment identification that enables precise deformable object tracking. Delivers robust detection capabilities even under heavy occlusions and self-intersections, utilizing non-intrusive tape-form markers for seamless application onto any surface.
Observations after Day 2 of porting OpenCV to Metal using AI: with a repo size of over 500mb, context management is king - you need guard rails in place for anything more complex than filters
Advanced research into vector-based social systems using 1408-dimensional embeddings for semantic search and personalized recommendations. Explores hybrid database architectures and RAG systems for intelligent content discovery and user matching.
Research Background
Professional Experience
CTO & Cofounder
Glyph Platforms
January 2023 - June 2025
Pioneered novel real-time computer vision pipeline for deformable surface tracking. Led research team and productization efforts, deploying to enterprise clients worldwide.
Research Scientist
Epson Canada
May 2018 - June 2025
Research and deployment of state-of-the-art deep learning technologies for industrial automation. Focus on object detection, pose estimation, and robotic manipulation systems.
Computer Vision Consultant
NSIX Vision Inc.
September 2018 - September 2024
Development and deployment of production vision systems for industrial automation. Achieved > 99.5% reliability with < 500ms cycle times across multiple complex applications.
Research Capabilities
Core Research Areas
- • Object pose estimation and 6D tracking
- • Robotic grasping and manipulation
- • Visual anomaly detection systems
- • Real-time augmented reality systems
- • Multi-view estimation and reconstruction
Technical Stack
Education
PhD, Computer Vision & Robotics
University of Toronto • 2025
Self-supervised Learning in Object Pose Estimation for Bin-Picking Tasks
BASc Engineering Science
University of Toronto • 2018 • 3.9/4.0
Minor in Robotics
Publications & Patents
Recent Publications
Novel approach for improving object pose estimation through targeted synthetic data generation based on pose and occlusion error analysis.
View PaperMulti-view approach for robust 6D object pose estimation in challenging industrial scenarios with occlusions and varying lighting conditions.
View PaperSelf-supervised approach for object pose estimation using active learning strategies and real sample capture techniques.
View PaperPatents & Innovation
Novel fiducial marker system capable of encoding data and surface characteristics under deformation and variable lighting conditions for AR applications.
View PatentsResearch Impact
- • Published in top-tier robotics conferences (ICRA, IROS, RA-L)
- • Patents deployed in production systems worldwide
- • 8+ international showroom deployments
- • Industrial systems with > 99.5% reliability
- • 6+ years of PyTorch research experience
- • Multiple publications in computer vision and robotics