DeepSeek's newly proposed Engram architecture introduces a fundamental shift in Large Language Models by decoupling static factual knowledge retrieval from dynamic reasoning. In this post, we explore how separating memory from compute could reshape the future memory hierarchy of LLMs.

In this article, VISTA Lab Researcher Ülkü Tuncer Küçüktaş analyzes the infrastructure implications of this architectural leap, highlighting how it elevates host DRAM and NVMe storage from mere system overhead to first-class contributors in future HPC clusters and AI training pipelines.

Building a scalable and flexible monitoring infrastructure requires moving beyond traditional raw data collection. In this post, we explore how to utilize open-source tools such as Prometheus, Grafana, and custom Python exporters to transform SNMP-based data into actionable time-series visualizations.

In this article, VISTA Lab Researcher Birol Şen explores the limitations of standard SNMP monitoring, while highlighting how custom exporters developed in our laboratory provide a centralized, proactive fault detection framework for our high-performance GPU servers and dynamic IoT data pipelines.

Identifying gastrointestinal conditions requires an understanding of both subtle local cues and broad global patterns. In this post, we explore the implementation of MedViT (Medical Vision Transformer), a hybrid architecture that overcomes the locality bias of pure CNNs and the computational intensity of ViTs to provide a robust framework for high-stakes medical decision-making.

In this article, VISTA Lab Researcher Furkan Kutan explores the complexities of diagnosing gastrointestinal conditions from abdominal radiographs, while highlighting how the hybrid MedViT architecture implemented in our laboratory provides a robust and interpretable framework for high-stakes clinical decision-making.

Technological superiority in defense and complex industrial systems is no longer measured solely by the manufacturing phase, but by the ability to keep these systems reliably operational over extended lifecycles. Recent discussions at the 21st NATO Life Cycle Management (LCM) Conference underscore the critical need to transition from traditional reactive maintenance to data-driven, AI-enabled predictive sustainment.

In this article, VISTA Lab Researcher Yarkın Sargın explores the importance of real-world operational profiling, while highlighting how the advanced deep learning and computer vision algorithms developed in our laboratory provide the technological foundation for this global sustainment vision.

We are thrilled to announce the official establishment of VISTA (Vision, Intelligence and Secure Technology Applications) Lab within the Department of Electrical and Electronics Engineering at Gazi University.

Our lab aims to bridge the gap between theoretical AI research and practical applications in critical sectors such as healthcare, energy, and defense. We are actively looking for motivated MSc and PhD students to join our team.