In this role, I collaborated on a pioneering venture, SITA Robotics, leveraging computer vision to develop algorithms for scanning the undersides of trucks, enhancing security measures against smuggled or hazardous goods. Employing YOLO models and OCR libraries, I orchestrated license plate detection and developed a stitching algorithm using, optimizing anomaly detection for swift identification.

Holmatro has been leading the market in rescue equipment for decades now, and with the release of their Pentheon series, they were looking to take their software development workflow to the next level. Their small team relied on third-party contracts and often did not know how to read their existing firmware scripts. They also lacked an in-house CI/CD pipeline. Holmatro hired me to fix their DevOps issues, help them get more sturdy with their software development workflow, and implement new features and fix existing bugs in their firmware.I spearheaded transformative initiatives. Resolving DevOps challenges, I established a Docker-based CI/CD pipeline on Bitbucket, standardized coding practices, and introduced vital firmware features. Swiftly delivering a ready image for the pipeline within two weeks, I facilitated an efficient workflow. Confidentiality constraints aside, my contributions involved implementing device behaviour, extending sensor scripts, bridging communication protocols, and enhancing testing capabilities, significantly fortifying their firmware.

- Led a team of 5 students through a year-long graduation project called Bicycle Driver Safety Suit.- We created a prototype of a portable rear-view camera for bicycles that detects dangerously approaching vehicles and sends live warnings to users.- Throughout the project, I was responsible for data collection, model training and evaluation, as well as hardware selection and integration.Technologies used: Python, Intel RealSense D435, IntelNCS2, Raspberry Pi, YOLOX, SORT tracking algorithm.* Our project won 5th place at the 2022 CENG Demo Day.senior.ceng.metu.edu.tr/2022/BDSS/

- Created a GPU management web app for deep learning projects, which runs them on environments in Docker containers enabling many users to develop onthe same GPU server at once with no conflict.- Developed and polished an RTSP streaming simulator.- Worked extensively on stereo-vision depth sensors and their uses in traffic surveillance systems, resulting in their deployment in multiple projects and applications. Data modelling and visualisation was done using NumPy, OpenCV, Flask, & Intel RealSense SDK 2.0.- Tested an emerging camera brand that uses Moviduis Myriad VPUs. The focus was on its performance in running vehicle detection YOLOX models.- Helped develop a stitching algorithm for convertingvideo frames of moving trucks to panoramic view.- Trained YOLOX models for multiple products.- Created a multi-class evaluation tool for computer vision model detections.

Built an MVP website for a 3D kitchen catalogue startup to help generate leads for German retailers.

I worked on the backend of a patient generation website which uses doctors’ input describing the desired case to generate a patient whose vitals match that description.