Computer Vision
Visual recognition systems for images and video, including object detection, segmentation, classification, custom dataset preparation, and model evaluation.
Computer Vision | Applied ML | Data Products
Meet Atalay
A data scientist and applied AI researcher with an MSc in Computer Science.
Develops machine learning solutions that transform raw data into models and applications for prediction, process automation, visual understanding, and decision making.
Published research focuses on computer vision and reinforcement learning. The work involves turning messy images, text, tables, and behavioral data into usable models for real applications.
- 4+
- Years in Data Science
- 33
- Public GitHub repos
- 6
- Hugging Face Spaces
- 2
- Published Papers
Core Work
Applied AI work across computer vision, automation, credit risk, recommendation systems, and academic research.
Machine Learning Products
Search systems, recommendation engines, scoring models, dashboards, and decision workflows built around real user needs.
Automation and OCR
Extract, validate, and structure information from documents, images, forms, and complex digital processes.
Publications
Machine learning and computer vision research with academic publications
Selected Work
Instance Segmentation
Double DQN
Reinforcement Learning Approach to Instance Segmentation Using Double DQN
Double DQN refinement integrated with Mask R CNN to improve instance segmentation masks on COCO 2017 without additional training epochs.
Published in IEEE ASYU 2025
Computer Vision
YOLO
ASL Detector YOLO
American Sign Language detection built with YOLO and deployed across Hugging Face and Android surfaces.
Publication
Detection
Budgerigar Gender Determination
Computer vision work for identifying budgerigar gender through cere detection and model comparison.
Published in UBMK 2024 Proceedings
Product
WebGPU
otoPILOT Ikinciel
AI car search for the Turkish used car market using Qwen3, browser inference, and arabam.com crawling.
Credit Risk
Tabular ML
Synthetic Istanbul Credit Risk
Credit risk estimation with non financial features and a privacy preserving synthetic Istanbul dataset.
100,000 synthetic resident records in the study
Civic Data
Notebook
IBB Education Aid
Data insight notebook for Istanbul Metropolitan Municipality scholarship aid data.
Public analysis project using IBB open data
Reinforcement Learning
VizDoom
DoomRL Baselines
Reinforcement learning baseline project for Doom with VizDoom and Stable Baselines 3.
Analytics
Tableau
National Currencies Per US Dollar
Interactive Tableau dashboard for comparing national currency values against the US dollar.
OCR
Selenium
Mersis Captcha Solver
Automation pipeline for solving captchas on the MERSIS portal using Selenium and Microsoft TrOCR.
Used in company verification workflows
Dataset
Captcha
MathCaptcha10k
Math captcha dataset and generation tooling designed for OCR experiments and captcha model training.
Dataset
Games
RAWG Games Dataset
Game metadata dataset prepared for analytics, recommendation, and ranking experiments.
CIFAR 10
Comparison
DNN vs DCNN
Image classification experiment comparing deep neural networks and convolutional models on CIFAR 10.
Publications
Advanced Computer Vision Techniques for Reliable Gender Determination in Budgerigars
Conference paper on object detection and deep learning for budgerigar gender recognition. Published in UBMK 2024 Proceedings.
DOI 10.1109/UBMK63289.2024.10773570
Reinforcement Learning Approach to Instance Segmentation Using Double DQN
Conference paper applying reinforcement learning to instance segmentation with a Double DQN approach.
DOI 10.1109/ASYU67174.2025.11208352