Kerem Demirtaş Email and Phone Number

• DLCP Split (Chief Developer)Increasing wafer utilization through efficient, fast, smart and robust clustering of microchips in a stochastic demand multichip package scenario for a variety of SKU stacks.◦ ApproachBinary tree based split design for clustering is implemented, where at each tree, an estimated bucketing penalty is computed as a performance indicator to be minimized. Then, optimization is done via a genetic algorithm that minimizes an objective fitness, most of which… Show more • DLCP Split (Chief Developer)Increasing wafer utilization through efficient, fast, smart and robust clustering of microchips in a stochastic demand multichip package scenario for a variety of SKU stacks.◦ ApproachBinary tree based split design for clustering is implemented, where at each tree, an estimated bucketing penalty is computed as a performance indicator to be minimized. Then, optimization is done via a genetic algorithm that minimizes an objective fitness, most of which consists of the estimated bucketing penalty giving the best clustering of chips to be held in the inventory.• Python 3 Migration (Owner)All the optimization and estimation packages were written in Python 2. A total of 16 packages needed to be migrated to Python 3 as Python 2 was reaching its end of life.◦ ApproachA smooth transition was necessary in order not to break the currently working packages. Servers were gradually updated to support both Python 2 and Python 3, and the packages were first transferred to be Python 3 compatible from pure Python 2. Then, Python 3 specific syntax and special methods were introduced into the packages.• Inventory Scenario Analysis – A Case Study (Developer)Helping the design process of an important product by a lifetime cost analysis for two scenarios using two different components that do the same job in terms of the end product, but, resulting in significantly different configurations. Analysis of pros and cons of favoring a certain type of component over the other and comparison of overall costs of the scenarios.◦ ApproachA case study was conducted with the data provided from several teams within the organization. Previously developed DLCP Split algorithm was implemented for efficient clustering of microchips and recommendation of optimal inventory holding policies. In the end, two scenarios turned out to differ by a margin of more than $100M in terms of capital tied to inventory on hand while reaching the same service level. Show less

• MIDAS: A Cyber Physical System for Proactive Traffic Management to Enhance Mobility and Sustainability (Funded by NSF) ◦ Problem Definition: Obtaining an overall snapshot of the whole road network through incomplete data from multiple sensors including aggregated loop detector observations, individual GPS information, instantaneous speeds, and images captured by dashboard cameras from limited amount of vehicles, and predicting the future state of the traffic for short, medium… Show more • MIDAS: A Cyber Physical System for Proactive Traffic Management to Enhance Mobility and Sustainability (Funded by NSF) ◦ Problem Definition: Obtaining an overall snapshot of the whole road network through incomplete data from multiple sensors including aggregated loop detector observations, individual GPS information, instantaneous speeds, and images captured by dashboard cameras from limited amount of vehicles, and predicting the future state of the traffic for short, medium and long term horizons. ◦ Methodology and Approach:Developed real-time traffic state estimation and prediction models combining big data, data fusion, forecasting and stochastic filtering algorithms with traffic flow models in Python using pandas, scipy, numpy and scikit-learn. • SOLARIS: Addressing Fidelity Between Meso and Micro Simulations to Evaluate Traffic Flows in Multiresolution Modeling (Funded by USDOT) ◦ Problem Definition:Evaluating and comparing microscopic and mesoscopic levels of traffic simulation to assess the fidelity gap between them with respect to different resolutions, accuracy and computational performance. ◦ Methodology and Approach:Developed a novel fast and accurate lane based mesoscopic discrete event traffic simulation framework in Python using pandas, simpy and networkx, which can reconstruct a ground truth scenario generated by the traffic microsimulation software VISSIM. Show less

Responsibilities include in-person lectures (twice a week), preparing and grading homeworks, projects, quizzes and exams for the IEE 376 Operations Research, Deterministic Approaches course.

Assisted Courses- Simulation- Project Management- Operations Research, Deterministic Methods- Network Flows

Responsibilities include in-person lectures (twice a week), preparing and grading homeworks, projects, quizzes and exams for the IEE 475 Simulation course.

Developed a simulation model in WITNESS to assess the performance of a car body shop, and validated the output of the assembly line with a discrete Markov chain model coded in Matlab and C.