Alibek Erkabayev
Senior Computer Vision Engineer · AR · 3D Vision · Real-Time Vision Systems · Edge AI
Istanbul, Türkiye · open to remote
About
Computer vision engineer with 6+ years of experience building real-time vision systems for industrial inspection, robotics, AR and edge AI — from research prototype to production on cloud and edge hardware. Expertise spans 2D/3D computer vision, object detection, tracking, visual localization and performance optimization on resource-constrained devices.
Experience
-
Senior Computer Vision Engineer — CDMVision
Designing and optimizing real-time computer vision systems for AR-based industrial inspection and dimensional validation.
-
Founder & Computer Vision Engineer — Bekai Vision
Independent CV & robotics studio for freelance projects between full-time roles.
- Optimized vision models for cloud and edge via quantization and TensorRT, achieving a 60% reduction in system latency.
- Automated scalable ETL and data-collection pipelines in Python, cutting dataset preparation from days to hours.
- Delivered technical consulting and backend architecture for AI-driven applications for global clients.
-
Web-AR.Studio
- Team Lead – Computer Vision Department: Led a cross-functional engineering team delivering production-ready AR experiences.
- Computer Vision Engineer: Engineered a real-time GPU-accelerated image pipeline in C++ and WebAssembly for high-precision positioning and mapping in web environments.
-
Machine Learning Engineer — UZAKTA Bilişim ve Tasarım Ltd. Şti.
Built a Flask-based management interface and high-speed socket-based communication between NVIDIA Jetson edge devices and client systems.
-
Researching Trainee, ROBOTIC Lab — Yıldız Technical University
Worked with FPGA platforms to design and test custom digital logic circuits.
Education
- MS, Computer Science — Konya Teknik Üniversitesi
- BS, Computer Science — Yıldız Technical University
Projects
-
C++/OpenCV slide-change detector (pybind11) plus a FastAPI pipeline that transcribes interview recordings (faster-whisper), OCRs slides (PaddleOCR) and grades answers with Gemini. Slide detection measured at ~500 fps @720p on a Ryzen 7 5800H (median of 3 runs).
-
C++17 people counter. YOLOv8 detection via ONNX Runtime, centroid tracking, IN/OUT line-crossing counts, SQLite logging and a Streamlit dashboard. Unit tests and CI.
-
Dense metric depth for UAV imagery. COLMAP sparse points plus Depth Anything V2 relative inverse depth, mapped to depth by a C++ polynomial least-squares fit. Unit tests and CI.
-
Docker and evo harness evaluating Basalt VIO on EuRoC MAV datasets. ATE RMSE 0.069 m on MH_01 and 0.054 m on V1_03.
-
Real-time webcam pipeline. pix2pix denoising plus optional TecoGAN video super-resolution, streamed to a virtual camera.
-
MiniNinja, a webcam hand-gesture fruit-slicing game. MediaPipe Hands tracks the index finger and slashes are detected from its velocity. Python prototype.
-
cuda-vision-lab
in developmentWork-in-progress CUDA lab for SLAM front-end kernels. The first stage (array ops) measured a 12.7-21.2x GPU speedup over CPU on an RTX 3060 Laptop.
-
NanoRec-CPP
in developmentC++20 screen recorder with a Dear ImGui UI and FFmpeg pipe encoding. Audio and transcription are planned.
Skills
- Perception
- object detection, segmentation, tracking, pose estimation, visual localization
- 3D vision
- depth, SLAM / VIO, feature matching
- Deployment
- TensorRT, ONNX, quantization, NVIDIA Jetson, real-time GPU pipelines
- AR & web
- C++ → WebAssembly, WebGL / OpenGL ES
- Languages & tools
- Python, C++, JavaScript, PyTorch, TensorFlow, OpenCV, ROS, Docker, FastAPI, Git