Вести

We are glad to announce the delivery of the Design of the AI Gateway document as part of the activities of SAIFA work package 3 entitled SAIFA Integrated HPC and AI runtime environment. 

The SAIFA AI Gateway shall provide a common user-facing access layer to the underlying capabilities including the project infrastructure, sector-specific applications, support services and links to European AI Factory resources. This arrangement reduces fragmentation and enables users from research, industry and public administration to follow structured paths from initial exploration to validated AI workflows.

National infrastructure and computing environment

SAIFA will use the PARADOX HPC cluster operated by IPB and the AI computing infrastructure operated by ITE for development, integration and preliminary testing. Both provide dedicated computing time and storage for project activities. The planned 100 Gbps connection between the IPB and ITE data centres, established with AMRES, supports efficient transfer and synchronisation of project data. Final validation and performance testing will also use the Greek and Italian AI Factory infrastructures in accordance with the access and interoperability arrangements developed in SAIFA project framework.

Existing scheduling and resource-management systems will be adapted to support priority-aware and workload-aware allocation. Users will submit jobs with defined CPU, GPU, memory and runtime needs, while the underlying schedulers remain responsible for queue management, backfilling, quality-of-service policies and fair use of the available capacity.

The computing environment will support containerised execution so that applications and their dependencies can be reproduced across compatible platforms. Kubernetes-based services and the existing cluster schedulers may be used for different workload types. Curated images for widely used AI frameworks will simplify onboarding and provide repeatable starting environments.

Advanced users may also register approved custom environments when a domain workflow requires specific libraries or tool versions. The approval and execution process must preserve security, reproducibility and compatibility with the selected computing backend.

AI Gateway layer

The AI Gateway will sit above the computing and data infrastructure as middleware: a unified access and integration layer that accepts user or application requests, applies common policies and metadata, and forwards the requests to the appropriate connected service or execution environment. It may expose applications, workflows, reusable tools, pre-trained models, AI-ready datasets, support and training resources without implementing their domain-specific logic.

Typical Gateway responsibilities include identity and access enforcement, catalogue and metadata management, request validation and routing, quota and policy checks, submission and status aggregation, monitoring, audit and provenance, consistent portal/API access and common rules for controlled sharing. Domain applications, workflow engines, model-serving systems, compute backends and authorised user-provided code remain responsible for scientific or industrial algorithms, data preparation, model execution, domain validation and interpretation of results.

Federation with European AI infrastructures

The Gateway may provide technical interoperability with Pharos, IT4LIA, the AI-on-Demand Platform and other EuroHPC AI Factories and Antennas. Depending on validated partner interfaces and governance conditions, possible integration functions include:

• Catalogue federation. Approved models, datasets, workflows and services may be made discoverable through connected catalogues while preserving origin, version, provenance and usage conditions.
• Metadata and access mapping. Catalogue metadata and access requirements may be translated between SAIFA and partner platforms.
• Validation evidence exchange. Approved testing and performance evidence may be transferred to support interoperability and service improvement.
• Controlled data and model exchange. Synthetic or other publishable assets may be transferred only where legal, ethical, licensing and policy conditions permit.
• Documentation access. Relevant technical documentation and tutorials may be surfaced through the Gateway.
• Training-service access. Relevant partner training resources may be linked where integrations are available.
• External service discovery. Relevant partner services may be surfaced without reproducing their internal functionality.

For more information about the Design of the AI Gateway document, please contact us at info@saifa.rs.

The Belgrade Bioinformatics Conference (BelBi 2026) is an annual international conference organized by several Serbian institutions, led by the Institute of Molecular Genetics and Genetic Engineering (IMGGE), University of Belgrade, as the main organizer. 

This year, the conference took place at the Metropol Palace Hotel in Belgrade, from June 8-10, 2026. On 10 June 2026, a special session was organized on Building Data Trust, where the SAIFA project was presented by prof. Zaharije Radivojević, from the School of Electrical Engineering, University of Belgrade. 

Additionally, the SAIFA representative from the Institute of Physics, prof. Marija Mitrović Dankulov was invited to take part at the BelBi conference opening panel. 

More about the event can be found on this link. 

On 11 and 12 June 2026, the Institute of Physics Belgrade organized and hosted the memorial conference In Memory of Antun Balaž – Insights into Complex Systems, dedicated to the memory of Dr. Antun Balaž, an outstanding Serbian physicist, visionary scientific leader, and key figure in the development of high-performance computing (HPC) in Serbia. 

The conference brought together nearly 70 researchers from leading institutions in Serbia and abroad, reflecting the broad scientific community that Antun Balaž helped build throughout his career. Colleagues, collaborators, students, and friends shared scientific insights and personal memories, highlighting not only his outstanding professional achievements but also his generosity, mentorship, and unwavering dedication to science. 

On the second day of the event, the SAIFA team organized the High-performance computing session, where representatives from the AI communities from the region (Italy, Greece, Bosnia and Herzegovina, Slovenia, North Macedonia and Bulgaria) had the opportunity to discuss the current challenges and future directions in the HPC ecosystem and the potential collaborations in the EuroHPC AI Factories initiative. 

More information about the event can be found at this link. 

On 14 May 2026, the kick-off meeting of the SAIFA – Serbian Artificial Intelligence Factory Antenna project was held at the Palace of Science in Belgrade, marking the official start of Serbia’s participation in the emerging European AI Factory ecosystem under the EuroHPC Joint Undertaking initiative.

SAIFA is led by the University of Belgrade School of Electrical Engineering and unites a national consortium of research and innovation organizations aimed at enhancing Serbia’s capabilities in artificial intelligence and AI-driven innovation. The project functions as the Serbian AI Factory Antenna, connecting national stakeholders with European AI Factory infrastructures, including direct partnerships with the Pharos and IT4LIA AI Factories.

The meeting brought together consortium members, institutional stakeholders, and national representatives to review the project's strategic positioning, management and governance frameworks, technical work packages, communication plans, and the roadmap for the initial project year.

Partners and Strategic Role of the Antenna

The SAIFA consortium comprises top Serbian institutions across AI, language and culture, engineering, energy, health, and environmental sciences. Through the antenna model promoted by EuroHPC JU, SAIFA will function as a national gateway to European AI and HPC ecosystems, enabling collaboration, technology transfer, training, and support for academia, public entities, and industry.

The project particularly focuses on integrating AI technologies with cutting-edge computing infrastructures and data-driven services. This approach aims to help Serbian researchers and innovators connect with larger European initiatives, including AI Factories, Centers of Excellence, European Digital Innovation Hubs, EOSC, and European data spaces.

HPC Serbia and SAIFA: Complementary Activities

With the launch of SAIFA, national efforts concerning High-Performance Computing (HPC) and High-Performance Data Analytics (HPDA) will still be coordinated and promoted via HPC Serbia. Meanwhile, activities centered on Artificial Intelligence, including AI applications, AI services, and the development of the AI ecosystem, will be managed through the SAIFA initiative.

This complementary approach will foster strong alignment between Serbia’s established HPC ecosystem and the quickly evolving European AI infrastructure landscape, supporting coordinated efforts in scientific research, innovation, industry adoption, and digital transformation.

Overview of the Kick-off Meeting

The agenda included a wide array of strategic and operational topics crucial for the project's successful launch. After opening remarks and institutional addresses, the consortium examined the project goals, impact, governance approaches, and communication strategies.

Dedicated sessions focused on:

• management structures and project governance,
• collaboration with Pharos and IT4LIA AI Factories,
• communication and dissemination strategies,
• development of the HPC & AI runtime environment and AI gateway,
• sector-specific AI applications,
• user support and training activities,
• planning of the first-year deliverables and milestones.

Special attention was given to developing the future AI gateway, ensuring interoperability with European infrastructures, onboarding domain-specific AI applications, and establishing support and training mechanisms for users and stakeholders.

The meeting concluded with an open discussion session and agreement on the immediate actions and coordination activities for the upcoming project period.

The SAIFA project is co-funded by the EuroHPC Joint Undertaking under Grant Agreement No. 101263280.