Salih Karagöllü
I work on national-scale financial infrastructure at TÜBİTAK BİLGEM, and I spend my research time on AI/ML, computer vision, and LLM systems.
GitHub ·
LinkedIn ·
email
Projects
Dual-model dental segmentation research
DentAI is a deep learning research project on dental image segmentation, from dataset preparation and mask generation to training and mobile deployment. I began with one model, then split the work into a clinical gingiva model and a mobile model after the data and use cases proved incompatible. I optimized the mobile side for pixel-accurate segmentation of teeth, caries, cavities, and cracks directly on-device, reaching 95.79% mean pixel accuracy, 90.10% weighted IoU, and about 2 FPS on a Galaxy S20 FE while beating the best prior detector by more than 18 points in primary accuracy. The work was supported by HAVELSAN AI JET BİGG and published in IEEE Xplore.
RAG assistant for live airport operations
I built AirportGPT during my internship at iGA Istanbul Airport, one of the world's busiest airports. It combines real-time flight data APIs with an LLM and uses RAG so answers stay grounded in live airport and flight information instead of model memory. It was built as a working prototype during my internship, with the team exploring it as a potential product.
XGPrice
Trend change forecasting in stocks
I built XGPrice around an XGBoost model with custom engineered time-series features and preprocessing pipelines. Over a one-year backtest, it produced about 54% alpha versus buy-and-hold. That result is from backtesting, not live trading, and any backtest-only evaluation carries overfitting risk.
Lane Centering
Autonomous driving graduation project
I built a lane detection and centering algorithm for autonomous driving using a real-time image processing pipeline. It comes from straightforward computer vision work, not a heavy model stack.
Experience
TÜBİTAK BİLGEM. I work on MYS (Mali Yönetim Sistemi), Turkey's centralized government financial management platform. It runs on a Java 21 and Spring Boot microservices stack with React, PostgreSQL, and Kafka, and the work is fully production-facing. I handle hotfixes, release deployments, Jenkins and ArgoCD pipeline operations, production monitoring with Pinpoint, and integrations with government e-services and banks. I also designed an end-to-end tax debt tracking flow across two microservices and fixed N+1 query bottlenecks in batch payment processing.
Scale AI. I contribute to RLHF workflows by reviewing and giving feedback on code outputs from state-of-the-art LLMs.
iGA Istanbul Airport. During my internship, I built AirportGPT with real-time flight data and RAG-grounded FAQ responses.
· 2nd place, TEKNOFEST International UAV Competition (Chief Software Engineer)
· Erasmus+ at Gdańsk University of Technology
· McKinsey Forward Program
· B.Sc. Computer Engineering, Gebze Technical University (3.03/4.00)
Java, Java 21, Spring Boot, Spring Framework, Python, JavaScript, TypeScript, React, Next.js, PostgreSQL, SQL, Kafka, Docker, Jenkins, ArgoCD, Pinpoint, Git, Bitbucket, GitHub, Microservices, REST API, OAuth2, CI/CD, DevOps, TensorFlow, PyTorch, Deep Learning, Machine Learning, Computer Vision, NLP, LLM, RAG, LangChain, YOLO, DeepLabV3, MobileNet, Quantization, Segmentation, CNNs, RLHF, XGBoost, Scikit-Learn, Pandas, Matplotlib, Data Augmentation, Data Preprocessing, Optuna, Jupyter, COCO, Object Detection, Image Processing, C, C++, HTML, CSS, Material UI, Shadcn, ASP.NET Core, Agile, Jira, Scrum, OOP, Spring Security, Hibernate, JPA, Kubernetes, AWS, EC2, Linux, Bash