Computer Vision Engineer
Eight years building real-time computer vision for embedded systems. I train and optimize models, then get them running inside the latency and memory budget of the hardware they ship on.
I have spent eight years building real-time embedded systems around image processing and deep learning, covering the whole model lifecycle: design, training, optimization and deployment across a range of hardware. Most of that work comes down to resource efficiency, taking a model that runs in a lab and making it run on an edge device, including size reductions of up to 95 percent on production systems.
I am the primary inventor on a granted US patent (US 12602797) for multi-camera 3D pose estimation using 3D joint heat maps. Alongside my professional work I build side projects end to end, most recently Fortune Snap, a cross-platform mobile app published on Google Play that runs a vision model over user-captured photos on a serverless backend.
Ask a question about my experience, or start with one of the suggestions below.
Experimental retrieval-augmented setup running a small edge model. Answers are generated only from the text of the CV, so they can be incomplete or out of date. The PDF CV is the authoritative version.
MSc, Cognitive Science
Middle East Technical University
2023
BSc, Electrical and Electronics Engineering
Middle East Technical University
2017
Python across the model lifecycle, from experimentation and training through to verification. C++ for deployment on edge devices, for integration into existing C and C++ codebases, and for system-level work such as core parameter processing and high-precision tracking. Matlab for numerical computing and mathematical modeling.
Object detection, semantic segmentation, 2D and 3D human pose estimation, anomaly detection, and radar spectrogram analysis. Synthetic data generation and optical simulation in Blender through the bpy API, alongside optical design, lens selection and camera calibration.
Getting trained models onto constrained hardware and keeping inference inside its latency and memory budget, including model size reductions of up to 95 percent on production systems.
Linux-first development across local and remote workspaces, versioned team collaboration, and automated annotation and data management pipelines built on V7 Darwin with custom Python tooling.
The full breakdown, with dates and employers, is in the CV.
Journal article
Journal of Medical Imaging 13(4), 045501, 2026
Adversarial evaluation of nine commercial vision-language models on 600 brain MRIs across roughly 27,000 inference calls, showing that OCR-readable text embedded in an image overrides pixel-level evidence.
Patent
US 12602797 · Granted April 14, 2026
Primary inventor of a method for multi-camera 3D pose estimation using 3D joint heat maps, addressing synchronization and occlusion in multi-sensor capture setups.
MSc thesis
Middle East Technical University, October 2023
EEG-based cognitive workload classification using deep learning over wavelet images. EfficientNet-B0 reached the highest accuracy, though performance varied enough across subjects to limit generalizability.
Side project, 2026 to present. Sole developer. Published on Google Play.
Cross-platform mobile app that runs a vision model over user-captured photos, backed by a serverless Firebase backend that handles inference, a credit system, and in-app purchases. Built to take my computer vision background end to end on mobile, from camera capture through to a published store release.
The full CV covers roles, dates and the projects behind each one.