Work Experience

---

• Member of a new product team. The tech stack for my role includes Python, Django, Datadog.

• Scaling Point of Contact for two critical services. Member of API Gateway team for Prime Video device traffic. Implemented improvements on the validation, authentication and decoration of millions of requests per second for Prime Video with low latency, high availability and low cost. My role was networking infrastructure heavy. The tech stack for my role included Java, TypeScript, Spring, AWS, AWS-CDK and DNS Configuration.

• Worked as part of the communications team. Developped solutions to optimize Udemy's communication pipelines using Python, Django & Kotlin. Implemented an internationalization solution that helped our translators to update and implement new features. Used a CI/CD pipeline integrated to github for AWS deployment. Increased the test coverage of our codebase by more than 10%. Monitored releases of new features through Datadog and created dashboards for continued insight.

• TÜBİTAK Intern Researcher Scholarship holder. Working with Pinar Yanardag from MIT on GAN inversion & latent space manipulation methods using Python, PyTorch & Tensorflow, such as finding meaningful directions in the latent space and latent space manipulation based on labeled directions found by supervised / unsupervised methods.

• Using Scala, implemented a Multivariate Gaussian Process Regressor that offers budget limitations and is online trainable resulting in significantly reduced runtime . This model is now deployed and it is being used by Tazi's main product. Implemented a data dashboard tool with React.

Education

---

Bogazici University

Information Retrieval, Applied Computer Networks, Database Systems, Machine Learning

CentraleSupélec

Cloud & Distributed Computing, Agile Management of Complex Projects, High Performance Simulation

Publications

---

Using the style space of the StyleGAN2 model to perform disentangled control of target attributes, we propose a method that does not require training additional models and directly debiases the GAN model output distribution by introducing an additional sparse layer into the matrix. I was the lead author of the paper and wrote it as an extension of my graduation thesis. The paper was published in ECCV 2022.

Projects

---

Lisan Language Learning App

Used Python, Django to develop a full-featured language learning app called Lisan, complete with a forum, courses, units, tips, lessons, word by word translations and question discussions. Lisan currently offers German and Bosnian courses for English speakers as well as an Italian course for Turkish speakers. The project was initially deployed with Heroku, but is now hosted on AWS. It makes use of Celery for periodic tasks.

CCTV mask usage detection

Collected data using OpenCV Labeled data with a simple labeling tool coded in tkinter. Created a Keras model to classify people into masked and non-masked. Collected data for several days to track changes in mask usage habits in Istanbul. Was contacted by the Istanbul Municipality R&D team for possible future collaborations.

Languages

---