As a software engineer at Ordinatrum, I develop innovative and engaging mobile applications using C#, Python, and Unity. I have over four years of experience in the field, including three and a half years as a mobile application developer at another company. I also completed two summer internships at Ordinatrum and A GRUP ŞİRKETLERİ, where I gained valuable insights into the industry and the latest technologies. I hold a bachelor's degree in computer engineering from Istanbul Technical University, where I graduated with high honors in 2019. My core competencies include problem-solving, creativity, collaboration, and communication. I am passionate about creating user-friendly and accessible solutions that meet the needs and expectations of diverse audiences. I am always eager to learn new skills and explore new challenges in the dynamic and fast-paced environment of mobile development. My goal is to contribute to Ordinatrum's mission of delivering high-quality and cutting-edge products that enhance the lives of millions of users around the world.
Led the development of Nutrition AI, designing core algorithms for real-time personalized nutritional recommendations. Managed and mentored a technical team, ensuring high-quality delivery and best engineering practices. Directed the development of the ARC Web Application, driving ICU workflow digitization. Coordinated cross-functional collaboration with clinicians and stakeholders. Oversaw project planning, code reviews, and technical decision-making to deliver scalable and impactful healthcare solutions.
Designed scalable and robust system architectures for the ARC Web Application and ICU decision support platforms. Architected solutions for real-time processing of high-volume medical data. Defined system structures for algorithm execution, prioritization engines, and data-driven workflows. Led architectural decisions for integrating 200+ medical device models, ensuring extensibility and maintainability. Ensured high performance, reliability, and interoperability across healthcare systems.
Developed core features for ICU-focused healthcare systems using C# and Python. Built and optimized custom algorithms for the Advanced Decision Support System to improve patient care efficiency. Implemented real-time data integration pipelines for processing patient and device data. Designed and trained an LSTM-based predictive model to forecast septic shock probabilities, enabling early intervention. Collaborated closely with healthcare professionals to ensure clinical accuracy and usability of solutions.
Worked on an ongoing .NET Desktop Application that gets the status (current, voltage, temperature, alarms etc.) of the robotic arm in real time and saves the status of the robot in an interval. Created an Raspberry Pi application that produces barcode for a product and reads the barcode to make sure it is correct product when packaging. Python programming language is used to create this application.
Created a website called HOBİ Panel using C# .NET MVC. Worked with Raspberry Pi and created a Linux Service to get data from Intensive Care Ventilator using Serial Communication and send the data to an API in Real Time. Python programming language is used.
Worked on the ARC web application, a transformative tool designed to digitize the work of ICU clinicians. My efforts were focused on ensuring the application streamlined the clinical workflows, enhanced data accuracy, and improved overall efficiency in the ICU environment. This project involved:
Developing and maintaining the application's core functionalities. Designing scalable and robust software architecture. Leading the technical team to deliver high-quality solutions. Collaborating with ICU clinicians to understand and meet their specific needs.
Developed a comprehensive web-based platform designed to enhance the efficiency and effectiveness of patient care in intensive care units (ICUs). This platform empowers healthcare professionals to:
Create and Manage Algorithms: Users can build custom algorithms tailored to specific patient needs and medical conditions. They can also add successors to these algorithms to ensure a seamless workflow.
Prioritization Rules for Patients: Implement prioritization rules to identify and address the most critical patient cases promptly, ensuring that high-risk patients receive immediate attention.
Calculation Formulas (Expressions): Designed a feature for users to create calculation formulas to generate new parameters when vital devices do not support certain measurements. This ensures comprehensive monitoring and analysis of patient data.
The platform leverages data produced by the ARC intensive care system, allowing users to utilize every type of data available to create algorithms, prioritization rules, or formulas.
Key Features:
Real-time Data Integration: Seamlessly integrates with the ARC intensive care system to fetch real-time data for accurate and timely decision-making.
Automated Notifications: When new data events occur, the Decision Support System (DSS) runs the created algorithms. If an algorithm produces a positive result, the system automatically sends notifications to doctors and nurses, facilitating prompt medical intervention.
User-friendly Interface: Intuitive design ensures that healthcare professionals can easily create, modify, and manage algorithms and rules without requiring extensive technical knowledge.
Impact:
Enhanced patient care through timely identification of critical cases. Improved efficiency in ICU operations by automating complex calculations and decision-making processes. Empowered healthcare professionals with tools to customize patient monitoring and care strategies.
Led the integration of various vital devices into the ARC system, encompassing patient side monitors, ventilators, infusion pumps, hemodialysis, and hemofiltration devices.
Key Responsibilities:
Established serial and network connections to receive raw data from a diverse range of vital medical devices. Developed and implemented data parsing protocols based on device-specific documentation to ensure accurate data interpretation. Integrated parsed data into the ARC system for real-time monitoring and analysis. Enhanced the ARC system's capabilities to display comprehensive and actionable patient data, thereby improving clinical decision-making.
Achievements:
Successfully integrated over 200 different models from more than 20 brands, significantly expanding the ARC system's compatibility with various medical devices. Improved the efficiency and accuracy of patient data monitoring, leading to better patient outcomes. Enabled healthcare professionals to receive timely notifications and alerts, improving response times and patient care quality.
Nutrition AI is an advanced healthcare project aimed at enhancing patient care in the ICU by providing precise nutritional recommendations. By leveraging real-time patient vital data, this AI-driven system predicts the optimal nutritional intake for each patient, including calorie count and the ideal balance of protein, carbohydrates, and fat.
Key Responsibilities:
Designed and implemented the core algorithms that analyze patient data and generate nutritional recommendations. Utilized advanced machine learning techniques to interpret patient vital signs and other health metrics, tailoring nutrition plans to individual needs. Worked closely with healthcare professionals to validate the nutritional models and ensure the recommendations meet clinical standards. Conducted thorough code reviews to maintain high quality and performance of the software.
In this project, I leveraged the MIMIC III dataset, which includes detailed clinical data from over 30,000 patients, to develop a predictive model for septic shock. The primary objective was to create an advanced machine learning model that can accurately predict the probability of a patient developing septic shock, enabling early intervention and improving patient outcomes.
Key Achievements:
Data Preprocessing: Cleaned and preprocessed a large-scale dataset comprising over 30,000 patient records, ensuring high data quality and integrity for model training.
Feature Engineering: Identified and engineered critical features relevant to septic shock prediction, including vital signs, lab results, and clinical notes.
Model Development: Designed and implemented a Long Short-Term Memory (LSTM) neural network model, leveraging its capability to capture temporal dependencies in time-series clinical data.
Model Training and Evaluation: Trained the LSTM model using a robust training pipeline and evaluated its performance using key metrics such as accuracy, precision, recall, and AUC-ROC.
Performance Optimization: Fine-tuned model hyperparameters and utilized techniques like dropout and regularization to prevent overfitting and enhance model generalization.
Results: Achieved significant predictive accuracy, demonstrating the model's potential to predict septic shock in patients with high reliability.
Impact:
The development of this LSTM model represents a critical step towards enhancing clinical decision-making in intensive care units (ICUs). By predicting septic shock probabilities with high accuracy, healthcare providers can initiate timely interventions, potentially saving lives and reducing the severity of septic shock cases.
DataAidKit is a dynamic web application designed to empower users with the ability to create customizable dashboards and generate comprehensive reports. This project leverages advanced data visualization techniques and user-friendly interfaces to facilitate data-driven decision-making.
Key Features:
Customizable Dashboards: Users can design and personalize their dashboards to display the most relevant data and KPIs.
Interactive Reports: The application enables the creation of detailed and interactive reports that can be easily shared and analyzed.
Data Integration: Seamless integration with various data sources ensures that users have access to real-time data for accurate reporting.
User-Friendly Interface: Intuitive design makes it accessible for users of all technical levels to navigate and utilize the application effectively.
Scalability: Built to handle large datasets, ensuring performance and reliability as data needs grow.
In this project, I created a continuous version of the Sokoban game, which traditionally involves discrete movement and actions. This new version requires a more complex approach for solving puzzles, making it an excellent testbed for advanced DRL algorithms.
I implemented two types of DRL agents:
DQN (Deep Q-Network): This algorithm uses a neural network to approximate the Q-value function, which predicts the expected future rewards for a given state-action pair. The agent learns to choose actions that maximize its cumulative reward. DDPG (Deep Deterministic Policy Gradient): This algorithm is suited for continuous action spaces. It combines the benefits of DQN with policy gradient methods, allowing the agent to learn both the Q-value function and the policy that determines the best action to take in any given state.
Achievements
Successfully developed a continuous version of Sokoban, providing a unique challenge for DRL algorithms. Trained and evaluated DQN and DDPG agents, analyzing their performance and learning efficiency in the Sokoban environment. Demonstrated significant improvements in agent performance over time, showcasing the potential of DRL in solving complex, continuous-action tasks.
Learnings and Insights
Gained deep insights into the workings of DQN and DDPG algorithms and their applicability to continuous action spaces. Enhanced proficiency in using Keras and TensorFlow for developing and training neural networks in a DRL context. Developed problem-solving skills and the ability to innovate in creating complex game environments and training intelligent agents to navigate them.
Future Work
Further refine the Sokoban environment to include more challenging puzzles and obstacles. Experiment with other advanced DRL algorithms such as PPO (Proximal Policy Optimization) and A3C (Asynchronous Advantage Actor-Critic). Explore the application of transfer learning to improve the training efficiency of DRL agents in similar environments.
TÜBİTAK-Sponsored Project (116E167)
During my Graduation Project, I had the opportunity to work on an exciting TÜBİTAK-sponsored project focusing on the application of deep learning methods for depth estimation from a single image, aimed at enhancing robot applications.
Key Responsibilities:
Developed a supervised learning agent to estimate both absolute and relative depth from a single image. Utilized a Fully Convolutional Residual Network (FCRN) to achieve accurate depth estimation. Collaborated with a multidisciplinary team to integrate the depth estimation model into robotic systems, improving their spatial awareness and interaction capabilities.
Achievements:
Successfully implemented the FCRN-based network, demonstrating significant improvements in depth estimation accuracy. Contributed to the overall project goals by providing innovative solutions and insights into the application of deep learning in robotics. Presented findings and methodologies in project reports and team meetings, facilitating knowledge sharing and further research in the field.
Skills Developed:
Deep Learning and Neural Networks
Computer Vision
Python and relevant deep learning libraries (PyTorch)
Data Analysis and Model Evaluation
Collaboration and Communication in a research environment
This project not only enhanced my technical skills but also provided a valuable experience in applying theoretical knowledge to practical problems, particularly in the realm of robotics and artificial intelligence.