- Product Unification Algorithm: Trendyol have more than 100 millionproducts. We need to determine whether we already carry this productin our catalog. Many products is duplicated this leads to a bad user ex-perience. Our system have two level. First retrive similar products usingsyntactic and semantic search. Syntactic search phase we use bm25 algo-rithm, in semantic search we developed siamese network with triplet loss.Among similar products we are calculating similarity using price, attributesand image information. Elasticsearch Open Distro is using this project forstore text, image vector and attributes. Image vectors are extracting fromCNN and this vector are using for image similarity and image search. OpenDistro supports approximate nearest neighbor algorithm (HSNW) for fastvector search.- Price comparision: Crawls e-commerce sites in the market and compares prices with product matching- Attribute extraction image and text- Detect Unwanted Images: Like alcohol, guns or nudity. Scraping datafrom Microsoft Bing for creating labelled dataset. Using EfficientB1 al-gorithm with Tensorflow. Developing API using fastapi framework anddeploying Kubernetes- Product Content Quality Score:It is a scoring system we developed tomeasure the content and quality of a product. It aims to make sure thatthe product contents on the site contain sufficient and diverse informationfor users. The quality of the product was scored by looking at featuressuch as image quality, key attribute occupancy rate, image count, productdescription- Seller NPS classification: Development of the module that automati-cally classifies the comments in the filled out by the suppliers every month

I am currently working / worked on:- CV: Deep learning state of the art networks YOLO, Mask R-CNN, AlphaPose- NLP: morphological analyze of Turkish Language, Word2Vec- Data Science: Linear Regression, Decision Trees, Ensemble methods, LSTM, FBProphet- Autonomous Marine Vehicle- Online Intelligent Proctoring Tool- Sprongo Intelligent Video Analytics (SIVA), is an intelligent platform that delivers automated angle analysis of movement. SIVA provides angular measurements of an athlete’s legs, hips, shoulders, and arms that can move within a video clip, eliminating the need for painstaking manual frame-by-frame measurements. With automated joint and posture detection, SIVA allows athletes and coaches to analyze posture and technique with ease.The system is designed with automated and low cost scaling in mind.A short commercial of the project:https://youtu.be/bxmT39M9FWgFrameworks: Pytorch, Tensorflow, Flask, Redis, MariaDBTools: Docker, Prometheus, Grafana, AWS- Forecasting/Predictions: Car Sales, Passenger and Freight Traffic for Airlines, Cryptocurrency Price, Credit AllocationPiribotA project for developing an Unmanned Surface Vehicle Platform that is now a part of SAHA Istanbul(www.sahaistanbul.org.tr). I took part in improvements of simulation environment for increasing the difficulty and reality of the task of maritime navigation and finding solutions for the same task through deep reinforcement learning.Tech Stack: Unity3d, PyTorch, RLLib, Unity ML AgentsKeywords: Deep Reinforcement Learning, Robot Navigation, Unmanned Surface Vehicles

Computer Vision: - Object Detection with Python (using OpenCV). - Object Recognition with Python (using and benchmarking Keras CNN, Darknet, Mask R-CNN, Tensorflow backend). - Data prep for a neural network, cropping various beer images (for supervised learning) from Geovision Group's field auditing image database. - Developing neural network for object detection with original image and object coordinates (x, y, w, h). - Developing beer-focused (as SKU (Stock Keeping Unit)) AI system. - Developing API for downloading images and uploading results with Python. Big Data: - Data Analysis and Manipulation with Python. - Forecasting with Python (using Keras LSTM).

Development of banking applications in the loans team

Doğal Dil İşleme projelerine destek verme.

Pl/Sql ve C# ile uygulamalar geliştirme