Turkcell (Turkey)

Autoscaling in cloud environments traditionally relies on threshold- based rules that struggle to adapt to complex, dynamic workloads. This paper presents a reinforcement learning approach to autoscaling applied to a real-world industrial application at Turkcell. We train a Deep Q-Network agent using actual production workload traces to make joint horizontal and vertical scaling decisions for the NSFW content filtering service within the AIHUB platform. Our experiments demonstrate that the agent successfully maintains service level objectives under both dynamic and static load conditions. Through a weighted reward mechanism, we show that operators can flexibly balance performance and cost, achieving either a 40% reduction in latency or a 32% reduction in resource costs depending on business priorities. The results provide evidence that RL-based autoscaling is viable for practical deployment in production cloud environments.

This paper comprehensively analyzes the technical design, architecture and performance dynamics of Turkcell's Central Product Catalog Management System (CPCM) and its React-based front-end RAF. CPCM has an important position in Turkcell's CRM ecosystem by providing integrated management of customer offers, prices, campaigns and product information in a central catalog. The study investigates the system's Java8 based backend architecture (EJB 3.1, JAX-WS, Hibernate), cache mechanisms powered by Oracle Coherence, Weblogic application server infrastructure and the user-centric design of the React-based RAF front-end. The performance analysis was performed over 1.46 million web service calls per day (CatalogServiceLite) and a 230 GB Oracle 19c database, evaluating the system's capacity to support high volumes of transactions. The findings demonstrate that effective cache management and modular architecture significantly improve system performance, but they also point out the necessity of optimization for complex entity searches and expensive logging procedures (100 GB log tables). Additionally, RAF's dynamic interface, which is supported by UseState/UseEffect plugins and React Router, enhances the user experience while pointing up possible areas for improvement in REST integrations and page load times. In addition to offering an industry reference model, this cutting-edge analysis offers crucial technical insights on how product catalog architecture should be organized in largescale telecom systems, particularly with regard to cache methods, version control, and data consistency.

The rapid proliferation of fintech applications has generated a massive volume of user-generated feedback, often in the form of unstructured textual reviews. Extracting actionable insights from such data is essential for improving user experience, product design, and service reliability. This study presents a comprehensive comparative evaluation of three topic modeling approaches-TF-IDF with KMeans clustering, Latent Dirichlet Allocation (LDA), and BERT-based clustering using SBERT embeddings-applied to 67,000 real-world reviews from a widely-used fintech mobile application. The analysis employs both intrinsic clustering metrics (Silhouette Score, Davies-Bouldin Index, Calinski-Harabasz Index) and interpretability-oriented heuristics (UMass Coherence, Jaccard Overlap) to assess model performance. Findings demonstrate that BERT + KMeans significantly outperforms traditional methods in capturing semantically rich and distinct topics, with practical implications for fintech stakeholders in product development, customer support, and strategic planning. The study also proposes a hybrid evaluation framework and emphasizes the importance of noise resilience, semantic precision, and scalability in modern NLP pipelines for financial feedback analysis. Limitations and future directions are discussed, including the integration of sentiment, intent recognition, and multilingual adaptations.

Recent developments in unmanned aerial vehicle (UAV) technology have given UAVs more processing and storage resources, paving the way for the concept of edge computing-enabled UAV networks. In this paper, we propose a cooperative multi-agent reinforcement learning-based computation offloading framework for a UAV swarm. Flying UAVs with missions can offload part of their tasks to neighboring UAVs or to fixed edge servers at terrestrial base stations. This approach reduces the total energy consumption of all devices during a core mission. Our framework helps UAVs form stable sequences of offloading decisions under uncertainties in a dynamic environment. This study demonstrates the superiority of the proposed deep Q-learning (DQN) algorithm to the existing Q-learning, heuristic, and random decision-making algorithms.

This study seeks to develop an innovative infrastructure for digital wallets that addresses the diverse requirements of both corporate and individual users. The main focus is on real-time monitoring of users' interactions on digital asset management software and generating instant actions through the analysis of this data. In this context, the main objectives of the wallet include providing local solutions as an alternative to high transaction fees, ensuring that assets are securely stored on local cloud systems with digital wallets, and developing applications that will meet the needs of Financial Technology companies and its business partners. Thus, it is aimed to increase the security, efficiency, and cost-effectiveness of digital asset transactions and to improve the user experience in digital wallet use in general. As a result, this study aims to provide a comprehensive solution that aims to offer significant advantages such as security, usability, and cost-effectiveness in the field of digital asset management and storage. It is expected that this study will contribute significantly to the expansion of the digital wallet ecosystem and to users' experiences in digital asset management.

The emergence of 6G networks paves the way for a new era of Artificial Intelligence (AI)-empowered integration of Internet of Things (IoT), autonomous mobile robots (AMRs) and extended reality (XR). This will revolutionise the ways users (e.g., humans, robots) interact with each other and their surroundings expanding their awareness beyond immediate boundaries. Driven by demand for more effective emergency response fuelled by adverse and increasing impacts of climate change, in this paper, we conceptually explore an architectural framework for intelligent emergency response services in 6G networks. The proposed architecture bridges an evolution of supporting technologies in 5G with future intelligent emergency response service requirements. We identify integration of 3D networking (3DN) and ambient IoT (A-IoT), empowered by native AI, network sensing and localisation, as key technologies for the proposed framework. We conclude the paper with discussion on current technology gaps and recommendations that would lead towards future 6G intelligent emergency response.

This study explores the potential and practical user prioritization-based network slicing solutions to mitigate communication challenges that may arise after natural disasters due to an explosive amount of data traffic requests according to the latest earthquake experience in Turkiye from the perspective of mobile network operators. Accordingly, this study investigates the development of more robust and resilient 6G communication systems by proposing the utilization of artificial intelligence/machine learning-empowered service management and orchestration applications in open radio access network intelligent controllers in both near-real time and non-real time, ensuring vital connectivity via managing service requests intelligently and assigning network slice functions in the control plane of core network accordingly during and after earthquakes.

Users’ desire for enhanced performance drive the inevitable technological progress on vertical applications in wireless communication systems. To meet these demands, researchers vigorously investigate potential 6G and beyond technologies and solutions. Reconfigurable intelligent surfaces (RISs) have risen in popularity and attracted the attention of academia as well as industry and seem to be a promising candidate for 6G technology. RISs are reflective metamaterials with many configurable elements consisting of pin diodes that have the ability to manipulate the impinging signals' properties, hence, enabling some sort of virtual control over the wireless channel. In this paper, we propose a novel over-the-air index modulation (IM) scheme through the use of a hybrid RIS with active and passive partitions to convey additional IM bits over-the-air. In addition we propose another modified scheme that allows the transmission for even more IM bits but at the expense of certain tradeoffs. First, we present the system model for both of the proposed schemes. Furthermore, comprehensive computer simulation results are provided and discussed presenting the superior bit error rate (BER) performance and additional benefits of the proposed systems compared to similar benchmarks.