Software development is moving from a code-centric activity toward an intent-driven engineering process, a shift often described as "vibe coding". While this transformation creates significant speed and productivity opportunities for the defense industry, working code is not always safe code.

In this article, VISTA Lab Researcher Aleyna Hafızoğlu explores the evolution from AI-assisted coding to agentic workflows. She details why vibe coding must be combined with human oversight, security testing, and risk-based governance layers—such as Rules, Skills, and the Model Context Protocol (MCP)—to build secure and reliable defense software.

In safety-critical domains such as avionics, AI solutions reliant on cloud computing fundamentally fail to meet core certification and safety requirements due to connection latency, data privacy breaches, and cybersecurity risks. In this post, we explore a high-integrity architectural framework to migrate mission-critical models safely to the edge.

In this article, VISTA Lab Researcher Ahmet Onur Kaplan demonstrates how a Multi-Tier Simulation and Certification-Oriented Hybrid Data strategy, combined with a deterministic Runtime Safety Monitor architecture, systematically closes the reality gap. This hybrid framework ensures low-latency inference while satisfying rigid standards like DO-178C and DO-331.

As AI workloads rapidly shift toward edge devices and embedded systems, the need for low-latency and energy-efficient inference accelerators has become critical. In this post, we explore why Field-Programmable Gate Arrays (FPGAs) are emerging as a powerful hardware platform for deep learning.

In this article, VISTA Lab Researcher Şeyma Karagözoğlu examines the hardware implementation of two major architectures: Convolutional Neural Networks (CNNs) and Transformers. She highlights the computational properties, hardware optimization techniques, and unique deployment challenges each architecture faces in the context of hardware-software co-design.

As artificial intelligence pushes toward the network edge, traditional neural networks often prove too resource-heavy for power-constrained IoT devices. In this post, we explore a biologically inspired solution to this challenge: Spiking Neural Networks (SNNs).

In this article, VISTA Lab Researcher Gökhan Kılıç details how the event-driven, sparse computational nature of SNNs makes them a perfect fit for the edge computing ecosystem. He also dives into the algorithmic training challenges and conversion methods paving the way for real-time, highly efficient distributed AI.

Deep learning systems for medical imaging often struggle when encountering real-world clinical data that differs from their training sets—a challenge known as domain shift. In this post, we explore how prompt-guided synthetic data generation can effectively address this issue in dental radiology.

In this article, VISTA Lab Alumnus Fatih Uysal demonstrates how combining expert knowledge with vision-language foundation models can generate controllable synthetic lesions. This approach significantly improves the robustness of AI detectors, particularly for small or rare abnormalities, paving the way for more reliable clinical AI.

Reinforcement Learning (RL) is driving a major transformation in autonomous decision-making in military aviation. In this post, we explore how RL enables systems to learn adaptive, optimal behaviors in complex and adversarial environments where traditional rule-based approaches fall short.

In this article, VISTA Lab Researcher Ozan Peker examines the strategic shift from "automated" flight to "autonomous" combat. He highlights real-world applications—ranging from superhuman air combat manoeuvring to resilient drone swarm coordination—and outlines the challenges ahead for deploying these systems safely.

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.