İlkay Atil

A new generation semantic text, image, and video search engine for technology research. Contrary to modern web search engines, the semantic context of images and videos are indexed and available for search, even without any tagged information or metadata. The system continuously gathers and semantically analyzes and indexes media from the internet. This required a complex system, combining many open-source tools and developing custom solutions to orchestrate more than 20 different tools and services.

I was responsible for the principal research, system design, and back-end development of the distributed services. I used machine learning to train models for image labeling and Docker, Kubernetes, and Ansible to deploy and manage multiple services inside a cluster. With the combined efforts of a team of researchers, we successfully built a semantic web search engine with more than one million indexed multimedia files on a prototype cluster.

A large scale multimedia archive containing more than 500,000 hours of TV broadcast video. The system has modules for automatically analyzing video for audio keyword detection, commercial detection, on-screen text recognition, and semantic analysis modules for content classification (violence, adult, and so on). I was responsible for developing semantic analysis modules with image processing and machine learning techniques. I also trained models on a large dataset and deployed them on a large cluster of servers (more than 80) while assuring model performance and reliability.