Scale & Collaborate: Next-Gen AI Architecture

Join us for a deep technical session exploring the cutting edge of AI infrastructure and agentic architecture. We will dive into two critical challenges facing modern architects: structuring effective multi-agent collaboration beyond simple tool use, and breaking the GPU infrastructure lock-in to scale inference across clouds.
Agenda:

• 17:30 - 17:35: Opening & Welcome – Alon Fliess, ZioNet CTO, AI Foundry MVP & RD
• 17:35 - 18:25: From Tools to Peers: Architecting Multi-Agent Ecosystems with MCP and A2A – Aviel Blumenfeld, AI Lead at IAI
• 18:25 - 18:35: Break
• 18:35 - 19:25: Beyond the GPU Shortage: Architecting Cross-Cloud AI Inference – Arthur Averbuch, Co-Founder at JellyCloud

Session 1: From Tools to Peers: Architecting Multi-Agent Ecosystems with MCP and A2A
Speaker: Aviel Blumenfeld, AI Lead at IAI (Israel Aerospace Industries)
As we transition from isolated LLM chains to mesh-like multi-agent ecosystems, a critical architectural confusion has emerged: the conflation of tool access (MCP) with agent collaboration (A2A). This lecture examines the distinct roles of these protocols and explains why encapsulating autonomous agents as stateless MCP "tools" creates a scalability bottleneck for complex, stateful workflows.
We will explore the complementary integration of the Model Context Protocol (MCP) to standardize internal resource connections (the "USB-C" of AI) and the Agent-to-Agent Protocol (A2A) to handle asynchronous delegation, multi-turn negotiation, and opaque execution between peers. Through a live Python architectural walkthrough using Pydantic AI and fasta2a, we will implement a reference architecture that layers these protocols effectively—proving how to maintain shared state and user transparency across a distributed agent network without vendor lock-in.
About the Speaker: Aviel is the AI Lead at IAI, driving AI innovation. Previously, he served as Head of AI at ZioNet & a CEO at compAI, where he directed AI architectural research. He is a seasoned entrepreneur with deep expertise in data science, AI solution design, and leadership.

Session 2: Beyond the GPU Shortage: Architecting Cross-Cloud AI Inference
Speaker: Arthur Averbuch, Co-Founder at JellyCloud
While Tier-1 providers face GPU scarcity and rigid reservations, thousands of cloud-grade NVIDIA GPUs sit idle across Tier-2 and sovereign providers. The challenge for the Cloud Architect isn't the hardware—it's the architectural "gravity" of networking, security, and lock-in.
In this technical deep dive, Arthur will demonstrate how to decouple AI inference from provider constraints. We’ll explore the engineering behind a cross-cloud Kubernetes control plane that treats fragmented providers as a single, fluid resource pool.
Technical Highlights:

• Intelligent Placement: Using AI-driven scheduling to route workloads based on real-time cost, latency, and hardware availability.
• Transparent Networking: Implementing a cross-cloud Service Mesh to secure workloads without public IPs or complex VPNs.
• Operational Parity: Extending your existing CI/CD pipelines (Azure DevOps, GitHub Actions, etc.) to remote GPU clusters with zero code changes.

About the Speaker: Arthur is a Co-Founder of JellyCloud, specializing in removing cloud boundaries. He is an experienced Enterprise and Cloud Architect with a background at AT&T and Amdocs, building solutions for some of the world's largest Telco companies. He is passionate about innovation, efficiency, and solving complex multi-cloud challenges.