Research Intern

• Contributed to Perceptual Graphics and Vision research under the guidance of Dr. Kaan Akşit, in collaboration with Adobe Research.
• Contributed to the development of "Odak," an open-source scientific computing framework designed for optical sciences, computer graphics, and visual perception.

R&D Software Engineer

• Key contributor to the development of an inverse rendering-based state-of-the-art fish biomass estimation project, leading to a peer-reviewed and published algorithm, and a highly accurate system with over 90% accuracy in biomass estimation. Employed software tools, including Mitsuba 3, PyTorch, and Python.
• Deployed and scaled the underwater fish biomass estimation software to the existing hybrid infrastructure.
• Developed 2D/3D visualization tools to validate and debug several machine learning pipelines.

Graduate Research & Teaching Assistant

• Actively participated in multiple research projects with a focus on inverse rendering and auto-white balance correction, resulting in several valuable publications.
• Contributed as a teaching assistant across a range of university courses, offering academic support and guidance to students.

Courses Assisted:

• Advanced C++ Programming, Spring 2023
• Data Structures and Algorithms, Spring 2023
• Programming Paradigms, Fall 2021 & 2022
• Agile Software Development, Spring 2022
• Object-Oriented Programming, Spring 2021 & Spring 2022

Academic Service:

• Reviewer, RCV Workshop ICCV
• Reviewer, ReScience C Journal

Undergraduate Research Assistant

• Developed and trained an autoencoder for real-world image denoising problem using PyTorch.
  GitHub Project Demo
• Worked on dataset generation using Blender3D.

• Analysed a ray tracing implementation in C++ using VTune profiler and detected hotspots.
• Parallelized the sequential ray tracing implementation using OpenMP.
• Benchmarked the sequential code with parallelised code in terms of effective cpu utilization, elapsed time and memory/cache utilization by using VTune and Valgrind profilers, observed up to 300 times performance improvement.

• Developed a system for visually impaired people which recognises a given banknote.
• The classification of the scanned banknote is processed using AWS Rekognition custom label service.

• Worked on a neural network architecture in PyTorch for single-label image classification that combines CNN and LSTM.
• Achieved better performance in terms of convergence speed by combining characteristics of both models.