I am responsible for developing and integrating new ML-based solutions with existing products. I also design and build the supporting data and MLOps infrastructure, ensuring scalability and efficiency. Additionally, I actively contribute to discussions with stakeholders, helping to shape the roadmap for ML products and align technical solutions with business objectives.

I have conducted cutting-edge research exploring the domain adaptation capabilities of latent diffusion models (LDMs) for downstream computer vision tasks, with a focus on semantic segmentation. This work has led to the introduction of novel domain-aware architectural components that actively encourage or discourage using domain cues to improve in-domain performance as well as cross-domain adaptation capabilities during fine-tuning. To further enhance the effectiveness of LDMs, I proposed new prompting techniques leveraging the cross-attention mechanism within diffusion models.These techniques enable improved in-domain and out-of-domain performance by effectively utilizing domain cues and semantic information. My research has demonstrated the potential for LDMs to adapt to various domains and tasks, with implications for real-world applications in computer vision.

I worked as a research assistant in the “Machine Learning in Sustainable Energy Systems” group at the University of Tübingen under the supervision of Dr. Nicole Ludwig. In collaboration with the Eurasia Institute of Earth Sciences Ocean and Atmospheric Sciences Group and the İTÜ Vision Lab at Istanbul Technical University, I worked on probabilistic air pollution forecasting. Paper is in preparation.

I assisted with the bachelor-level course ”INF3151 - Introduction to Machine Learning” by grading weekly student assignments and holding practice sessions every week to help students with theoretical and practical exercises.

As an undergraduate research assistant, I worked under the direct supervision of Dr. Gözde Ünal. I had the opportunity to work on industrial collaborations, which gave me firsthand experience working with messy real-world data and using post-hoc explainability techniques. My work formed the basis of my graduation project, titled “Development of Machine Learning Based Predictive Models for Assistive Reproductive Technology”.

I collaborated with domain experts to perform data organization and cleansing. Additionally, I developed artificial intelligence models for treatment purposes. To interpret the model outputs, I employed interpretability techniques. Furthermore, I deployed the models into production using Ray Serve and created APIs for them. I utilized GoF design patterns in my work and conducted unit and integration tests to ensure the system's functionality.

From April to October 2020, I worked as a research intern at Basira Lab under the supervision of Dr. Islem Rekik and graduate student Ahmed Nebli. During my internship, I collaborated with Nebli to develop and train a cascaded graph convolutional network system. This system uses multiple models, where each model takes input from the previous one in a sequential manner. Our goal was to predict changes in signal communication between brain connectomes in the future, which could be a possible indicator of dementia. Our work was published at the MICCAI PRIME Workshop, where we were awarded the Best Paper Award.

During my internship, I conducted a comprehensive literature review on General-Purpose Graphics Processing Units (GPGPU). I explored relevant research papers and academic sources to gain a deep understanding of the subject matter. Additionally, I actively participated in weekly project meetings by presenting and discussing papers that offered valuable insights and potential contributions to the ongoing project. Through this experience, I honed my skills in effectively reading and extracting key information from academic papers, enabling me to stay up-to-date with the latest developments in the field.