tanırım
About
Numan Bekir Özçelik

Numan BekirÖzçelik

AI & Computer Vision Engineer

I was introduced to computer vision while studying Electrical and Electronics Engineering. In the last two years of my studies I worked in the AI team of a smart city company, preparing datasets from traffic camera footage and training YOLO models. My graduation project was a controller that manages traffic lights using camera detections.

Along the way I saw that the number a model shows during training does not always tell how it behaves on real footage. That is why I set aside separate time for measurement in my projects: I repeat the evaluation with a tool I wrote myself and share results together with the conditions they were taken under.

After graduating I worked as a SCADA engineer, then built web projects and set up my own courier marketplace. Working across different fields taught me to think about the technical side of a problem together with the side that reaches the user. Today AI is at the center of my work. Alongside computer vision, I am also interested in language models and other areas of AI.

Experience

Founder & Software Developer

Mar 2025 - Jan 2026

Kuryevar® · Sole proprietorship

  • Motorcycle and car courier marketplace: map-based ordering, pricing and courier assignment, payments and operations management, with business and admin panels.
  • Custom projects: barcode-based sales and inventory software, corporate websites.

SCADA Engineer

Mar 2024 - Sep 2024

Boğaziçi Elektrik Dağıtım A.Ş. · Istanbul

  • Diagnosed communication faults of ABB and Siemens RTUs over IEC 60870-5-104, configured RTUs with Cisco and Huawei routers, and designed and configured SCADA screens with ABB PSE and PED500.

Software Support Specialist (AI Department)

Jul 2021 - Jul 2023

Ekin Smart City · Ankara

  • Prepared datasets from real traffic camera footage: filtering, augmentation with OpenCV, labeling with CVAT, train/validation/test splits.
  • Trained YOLO object detection models on GPU with CUDA and cuDNN, on Linux and Windows.
  • Validated trained models in real time on RTSP streams.
  • Automated data processing and file management with Python scripts.

Other Projects

AI-Powered Smart Intersection

Simulation of an adaptive traffic signal controller driven by YOLO11n detections from an overhead camera. Compared with conventional traffic lights, it reduced average vehicle wait in 21 of 22 scenarios (22% at the median), and wait for emergency vehicles such as ambulances dropped from 18-64 s to 0.3-23 s.

Object Recognition and Web Lookup

CPU-only app that identifies the object in a photo and gathers sourced information about it from the web. Combines zero-shot CLIP with a linear classifier trained on CLIP features: 93.5% validation accuracy on Caltech-101, against 86% for zero-shot alone.

Education

B.Sc. Electrical and Electronics Engineering

2019 - 2023

Haliç University · Istanbul

Graduation project: AI-Powered Smart Intersection.

Skills

Computer vision
YOLO, OpenCV, PyTorch, CUDA, CLIP, MediaPipe, CVAT, RTSP
Programming
Python, TypeScript, SQL
Tools and frameworks
Linux, Docker, Git, Next.js, NestJS, PostgreSQL, PyQt5
Industrial
SCADA, IEC 60870-5-104