o Lead researcher for developing rapid ML training systems for high data generating sources, including APS-U at Argonne National Lab and LCLS-II at Stanford Linear Accelerator Center.o Responsible for collaborating with the domain scientist to gather information regarding the ML training pipeline requirements for robust model training. o I use Pytorch and Tensorflow for training ML models on state-of-the-art HPC systems, including the Argonne Leadership Computing Facility (ALCF)o Developed a ML framework for training ML model at a rapid rate for data Linac Coherent Light Source (LCLS-II) and Advanced Photon Source Upgrade (APS-U) that generate data at Gb/s.o I collaborated with other Scientists at ANL and employed a service for automatic data annotation and model recommendation service to accelerate deep neural network training with a focus on ML-based scientific applications.o Our proposed data service achieves 100x speedup in terms of data labeling compared to the legacy system.o Our model service achieved up to 200x improvement in training speed.

• Performance Optimization of TensorFlow Servingo A novel performance tool for scalability estimation of TensorFlow serving. o A scalability optimizer that efficiently searches and finds the optimal configuration for the TensorFlow serving system to minimize response time, resource utilization, and meeting SLO.• Optimal use of Burstable Instances of Amazon EC2 (Spring 2017-Spring 2018)o Proposed CEDULE a workload intensity and SLO aware scheduling framework for amazon EC2 burstable instances.o Resource allocation was done through a highly accurate prediction model based on quantile regression with average prediction errors range from 1% to 15%.o The effectiveness of CEDULE is verified as it can triple the efficiency of spare resources while meeting stringent SLOs.• Teaching Experience:o CPE 301 Embedded System Design (Fall 2018)o CS 201 Computer Organization (Spring 2018)o CPE 301 Embedded System Design (Fall 2017)

• Workload Scheduling of Distributed Data Centres: o Proposed and developed a workload scheduler for distributed data centres to optimize the operational cost for cloud service providers and cloud service consumers. o In order to minimize the electricity cost temporal and spatial variations in electricity prices are considered. o Cooling cost is minimized by taking into consideration the temperature variations within data centres and across different locations. o Three different workload scheduling techniques are proposed taking into consideration different workload types. o The proposed scheduler is implemented in Java using CloudSim simulator by adding time-varying characteristics to data centres.• Teaching Assistant at KOÇ University (Fall 2014 to Fall 2016)o COMP 132 Advance Programming (Spring 2016)o COMP 303 Computer Architecture (Fall 2015)o EQUR 101 Quantitative Reasoning Using Computers (Spring 2015)o COMP 130 Introduction to Programming (Fall 2014)

Work experience includes:• Development of new modules in a web-based application as per user requirements.• Validation of system functionality, incorporation of client's requirements for new modules and reporting of results.• Tested and develop new models for reports in VBA, from time to time to remove any inconsistencies. • Responsible for Unit testing and System testing of the reports based on the specification. • Requirement gathering, End to end Stakeholder management, Client communication, Reports Delivery, Resource management & Task allocation