Istanbul, Istanbul, Türkiye
Managed and worked with a team consisting of data scientists, data engineers & interns working on: - Suspicious Transaction Detection: Detection of suspicious money-laundering activities via Supervised ML models - Graph Based Detection: Pagerank inspired link analysis algorithm to detect suspicious entities - False Positive Reduction: Elimination of false alerts of sanctioned & suspicious entities generated by existing rule based AML tools via ML-models based on expert compliance officer behavior & n-gram models - Active Learning: Improvement of the ML models via Active Learning techniques with the participation of the compliance officers - Federated Learning: Application of the paradigm to the compliance & regtech domain - Professional DS Services: Providing data Health Check reports & rule optimization tuning for existing AML products - Credit Card Fraud Detection: Anomaly detection ensemble & LLM based Explainable Artificial Intelligence (XAI) Module - Market Manipulation Detection: Modular scenario based alert generation infrastructure for investment companies - Keywords: Supervised Learning, Link Analysis, Active Learning, Kneser-Ney Models, Simulated Annealing, Object Oriented Programming, Federated Learning, LLM
Optimization & Automation: Modelling and implementation of algorithms in the User Acquisition context for hyper-casual and casual game marketing campaigns on different ad networks (Unity, Ironsource, Mintegral, Facebook, Applovin). - Arpu & LTV Prediction - Retention Estimation Collaboration with the Monetization team to optimize monetization processes and improving game revenue metrics eCPMs, A/B Testing for both UA & Monetization Keywords: Statistics, Object Oriented Programming, Metaflow, AWS
Marvin, the AI-powered marketing assistant: Campaign performance optimization in the Facebook marketing ecosystem: - Budget allocation among ad-sets or campaigns - Bid Management - Bid Multipliers optimization - Conversion Prediction - AB Testing Project Galaxy: - Complete overhaul of the existing code-base to build a modular, scalable data science framework ML Supported Audience Creation: - Model performance improvement research & implementation Keywords: Reinforcement Learning, Bayesian Statistics, Object Oriented Programming, Supervised Learning, TF, Kubeflow
As of 2020, the Adphorus brand's been sunsetted and merged into Sojern.
Founding member of the new Data Science Lab Development of several decision support products: - Location+ : A location analytics solution giving suggestions on the next best locations to deploy ATMs (or similar entities) through evaluation of GIS Information. - Predictive Maintenance : Component based ATM failure prediction. - Credit Risk Analytics : Project on customer default probability detection & limit modification decision support - Task+ Addendum : Forecasting of incoming calls for call centers. Participation to Microsoft’s Machine Learning Hackfest in Milan (December 2016): Member of the Turkish Team in the event, working on the improvement of transaction forecasting in ATMs. Tools: Artificial Neural Networks, Decision Trees (CART, Random Forests), SVM, Boosting (Adaboost , Gradient Boosting Machine), PCA, Feature Selection (Mutual Information, Pearson Correlation etc..), Logistic Regression, Linear Regression
Modelling and development of several optimization-based decision support products: • Task+ : A workforce and resource planning tool which uses advanced simulation and optimization techniques to determine the optimal resource configuration in queue-based service environments. • Cash+ : Cash management optimization product which aims to minimize transfer and operational costs and idle cash, maximizing total liquidity and customer satisfaction. • Reload+ : A supply chain planning and route optimization tool which aims to solve different types of complex problems in logistics and various service industries. Participating to the development of the core P1M1 software infrastructure: • P1M1 Core Framework: The framework upon which the modular and intuitive P1M1 products are constructed. • P1M1 Core Engine: The engine incorporating various optimization engines developed in-house.