Computer Vision & AI Software Engineer
Ph.D. · M.Sc. · B.Sc.
SLAM · 3D Perception · Deep Learning · Estimation · Robotics
Gayan is a computer vision and deep learning software engineer with 8+ years of ML-based real-time software development experience, and holds a Ph.D. and M.Sc. in Electrical & Software Engineering from the University of Calgary. His work is grounded in Bayesian estimation, SLAM and 3D perception, with a track record of designing and optimizing AI-driven vision systems for intelligent video analytics — spanning classification, object detection, depth prediction and full 3D reconstruction.
He works fluently in C++ and Python to build real-time, scalable solutions for complex computational problems, with particular depth in custom CUDA-accelerated development and GPU-based recursive probabilistic estimators. Outside contract work, he has designed and led robotics projects — including an interactive receptionist robot for the City of Calgary — and shares much of his research on GitHub and YouTube.
Schulich School of Engineering, University of Calgary
Estimation theory, SLAM, deep learning and AI-generated 3D.
Schulich School of Engineering, University of Calgary
Grade: 6A+. Thesis: RGB-D SLAM for Autonomous Robots.
University of Moratuwa, Sri Lanka
Second Class Upper Division.
Complex System Inc., Calgary, AB, Canada · Mar 2018 – Sep 2025
Leads video analytics projects end to end — research, real-time implementation and deployment — optimized for hardware constraints, performance and reliability.
Arthur C. Clarke Institute for Modern Technologies, Colombo · May – Aug 2015
Reverse-engineered and redesigned a low-cost CCD tea colour sorter for the Sri Lankan tea industry, covering mechanical and electronic design and manufacturing oversight, for the Ministry of Technology Research and Atomic Energy.
S-lon Lanka (Pvt) Ltd, The Capital Maharaja Organisation, Sadalankawa, Sri Lanka · Dec 2014 – May 2015
Responsible for electrical maintenance of PVC pipe manufacturing machinery.
Power Migration Partners UK Ltd, Greater Edinburgh Area (Remote) · Apr 2013 – May 2015
Delivered embedded electronics and controls solutions for the UK renewable energy and industrial automation sectors — including sensor-driven joystick control for the Skoog music instrument, embedded programming for hybrid-energy inverter controller boards, and precision microfluidic pump driver design.
Dipped Products PLC · Mar – Oct 2011
Vizu Private Ltd · Oct – Dec 2010
Solar-powered Arm Cortex-M55/U55 edge device classifying user-defined urban sound events at over 95% accuracy, using quantization-aware training and structured pruning, as part of a multi-modal smart-city monitoring platform.
Led a team of four building a monocular 3D obstacle-detection system for the ISS robotic arm, combining YOLO detection with a custom depth model quantized for real-time 5 FPS edge deployment on Nvidia Jetson.
Lead designer and developer of an interactive robot's full mechanical, electronic and software stack — ROS-based, with custom motion, vision and voice subsystems — built to greet and assist visitors at Calgary's municipal building.
A Python/TensorFlow implementation of the Dual Convolution Vision Transformer, built from PhD research, for unsupervised, metric-scale depth prediction from single images.
A multi-threaded C++/GTK SLAM system for Linux, with dedicated threads for depth prediction, local mapping, sparse bundle adjustment and loop closing, plus parallel keypoint matching via RANSAC and EM.
A hexacopter drone built from off-the-shelf components with edge-AI processing, RGB/monochrome cameras and radar, alongside a joystick-controlled GUI for lab experiments.
A custom wide-baseline stereo rig — three monochrome global-shutter cameras plus one high-resolution colour camera — with IMU/GPS ground truth, collected across Calgary terrain for SLAM and depth research.
Multi-view RGB and RGB-D crowd analytics, fusing a fine-tuned YOLO detector with a custom stereo-based 3D human detector, scaled to 50 edge nodes running at 10 FPS for real-time people counting and service-time tracking.
Enhanced outdoor spatial mapping using Intel RealSense depth cameras. Improved localization and 3D reconstruction accuracy by fusing dense point clouds into a pose graph optimization pipeline for robust environment modeling.
Ported a robot's software from Windows to ROS-Kinetic, then built control software for sensor fusion, target detection, path planning and trajectory navigation, integrating a laser scanner and RGB-D camera.
Led mechanical and electronic design of a low-cost CCD-based tea colour sorter for the Sri Lankan tea industry, including manufacturing oversight.
Reverse-engineered and redesigned an HPLC pump for super-heated extraction research at the Faculty of Applied Science, University of Kelaniya.
Mechanical, electronics and control-system design for a self-balancing robot, using sensor fusion of gyroscope and accelerometer data for real-time stability.
Competition robot built with a two-person team to autonomously detect and transport bricks to a goal location, covering mechanical, sensor and control design.
One of three representing Sri Lanka; led mechanical design, fabrication and system integration of two fully/semi-automated pyramid-building robots.
Analyzed magnetic-sensor joystick output patterns and built a multivariate Gaussian model to estimate pressure-point location for the Skoog music instrument.
Embedded programming and I/O calibration on the F28M35M52C microcontroller for a hybrid-energy inverter controller board, with extensive state-space testing.
Re-engineered a sinusoidal-drive microfluidic pump circuit, adding constant-current control and a redesigned PCB layout for stability and reduced interference.
Designed a wireless RF production-line monitoring system, tracking per-operator task times across hundreds of input nodes for a master display.
Built a code-based RF communication system for a steel and concrete wire plant, enabling reliable status updates in a high dust/noise manufacturing environment.
15+ years across C, C++ and Python, plus microcontroller programming. Proficient in OpenCV, OpenGL and GTK for real-time, sensor-driven software pipelines.
ROS, PyTorch, TensorFlow, Keras Functional API, ONNX, TensorRT, MATLAB.
CNNs, Vision Transformers, supervised and unsupervised learning, object detection and tracking, model quantization and pruning, multi-task learning, large-model distillation.
Custom CUDA development, GPU-accelerated recursive probabilistic estimators, TensorRT/ONNX deployment, quantization-aware training, structured pruning.
Estimation theory, SLAM and 3D perception (sensor-based and deep learning-based generative 3D), monocular and stereo vision, multi-view geometry, optical flow and image registration.
3D CAD, circuit and PCB design, sensors, actuators and microcontrollers, low-level real-time software.
Feel free to reach out for collaborations, roles, or questions about any of the projects above.