LinkedIn
GitHub
>hello friends, welcome!
Machine Vision Specialist | Former Geospatial Vision Researcher
I specialize in machine vision and geospatial computer vision, building deep learning models that analyze satellite imagery, industrial production systems, and detect changes in Earth observation data. I recently completed geospatial research at the GeoVision Lab, University of Windsor, where I developed and trained a Siamese Attention U-Net model using TensorFlow/Keras for change detection tasks on geospatial image pairs. Currently, I'm focused on industrial vision inspection, working as a Vision System Inspection Technician where I monitor high-precision defect detection systems for battery manufacturing using advanced data analysis frameworks.
My expertise bridges computer vision research and production-ready web applications. I've integrated state-of-the-art vision models (YOLOv8, YOLO11, MediaPipe, RetinaFace) into full-stack systems built with React, Next.js, and Node.js, creating intelligent applications that process visual data in real-time.
From winning 3rd place at WinHacks 2024 with a geospatial energy visualization app to participating in HackTheNorth 2025 (Canada's premier hackathon), I thrive at the intersection of deep learning, satellite imagery, and scalable web development. Whether it's training attention-based architectures, building embedded vision systems on ESP32, or architecting full-stack AI applications—I turn research into reality.
💡 Passionate about: Change detection, attention mechanisms, geospatial analysis, and deploying vision AI at scale.
Education
Bachelor of Computer Science
University of Windsor
Professional Experience
Developed a patching system for creating splits of any satellite image, tested using QGIS render tech, and built a web scraping system to extract data in a structured way from database websites. Trained a Siamese Attention-UNet model using TensorFlow/Keras on geospatial image pairs to detect visual similarity for change detection tasks, optimizing training under constrained GPU resources (NVIDIA GTX 1080) by tuning batch size, input resolution, and memory growth settings to avoid OOM errors — then scaled the full training run to an NVIDIA RTX 4080 (Ada Lovelace), cutting training time to under an hour on heavy datasets.
Researched and evaluated frontal face detection techniques by integrating YOLOv8n with MTCNN and YOLO11 with RetinaFace, achieving improved detection accuracy and frame-level face counts across ML experiments. Conducted performance benchmarking and model comparison to assess inference speed, scalability, and detection reliability for computer vision applications.
Mentored 200+ students in software engineering fundamentals including Data Structures and Algorithms (DSA), problem-solving techniques, and best practices for writing secure, maintainable, and performant code.
Tech Stack
Languages
AI / ML / Computer Vision
Web & Full-Stack
DevOps & Cloud
Geospatial & GIS
Skills
Languages
AI & ML
Web & Full-Stack
Embedded & IoT
DevOps & Cloud
Geospatial & GIS
Featured Projects
tour2toronto
August 2026Built an interactive live map dashboard surfacing real-time events happening across downtown Toronto, consolidating festivals, markets, and concerts scattered across multiple platforms into one geographically-aware interface. Implemented time-aware event pins (live, starting soon, upcoming, ended), category filtering, and an AI-powered assistant for event recommendations grounded in real pinned data, using the Ticketmaster Discovery API and the City of Toronto open-data festivals feed.
YOLOv8 TensorRT Inference Benchmark
July 2026Benchmarked YOLOv8n inference on an NVIDIA Tesla T4 GPU, comparing native PyTorch against a TensorRT-exported engine, achieving 74.3 FPS (13.47 ms/frame) versus 67.3 FPS with PyTorch — a 1.10x speedup. Built an end-to-end benchmarking pipeline with Ultralytics, exporting to an optimized TensorRT engine and generating automated FPS/latency comparison charts.
AirBridge
February 2026Built an ESP32-based wireless middleware that enables direct photo and video transfer from smartphone to laptop using Wi-Fi AP mode and embedded HTTP server. Implemented firmware in ESP-IDF with focus on networking, memory constraints, and reliable data streaming on resource-limited hardware.
Ultimate Posture Watcher
uOttaHacks 2026 | January 2026Built a privacy-first, real-time posture coach using webcam input, MediaPipe Pose (WASM), and ergonomic angle detection. Designed configurable alert system with beeps and cooldown logic, plus session analytics tracking posture time and worst angles.
AI-Powered 3D Virtual Classroom
HackTheNorth 2025 | September 2025Contributed to building an immersive 3D classroom in Three.js with interactive areas for collaborative learning. Integrated Cohere's Command A Vision Model to process handwritten notes, achieving rapid LaTeX conversion in 3 seconds.
Face Detection & Clustering
September 2024Implemented face detection and counting system using Autoencoder for feature extraction and YOLO models. Applied KNN clustering to group similar faces with recorded accuracy metrics for evaluation.
Global Electricity Resources App
WinHacks 2024 | February 2024 🥉Assisted in API integration using React, Geocodify, and Electricity Map for real-time data visualization. Collaborated in a team of 4 to develop a resource solution, securing 3rd place in the competition.
Leadership & Community Impact
ACM Club Event Assistant
Spearheaded event operations for the university's premier computer science organization, orchestrating 10+ high-impact technology events annually that consistently achieved 70%+ student participation. Directed cross-functional teams across venue logistics, technical setup, and community outreach—building a thriving tech ecosystem that connected hundreds of students with cutting-edge CS concepts and industry trends.
Additional Projects
Docker Containerization & CI/CD
August 2023 – PresentOrchestrated Docker containers for Python web applications, ensuring seamless deployment across environments. Implemented robust CI/CD workflow using GitHub Actions, automating build and deployment processes for web apps and data structure implementations.
NASA Space Apps Hackathon
Windsor Edition | September 2023Cleaned and converted database to JSON format using Streamlit. Implemented user-friendly search functionality for data retrieval and automated machine-independent digital resume web page portfolio.
MINI ARCADE
Stick Arena
Take on the training bot in a quick physics-inspired stick fight.
MINI ARCADE
Comet Drift
Steer a comet through an asteroid field, gather signal crystals, and survive the storm.