| Page 160 | Kisaco Research

Train and Deploy High-Performing AI Model Development at Scale

In this session, we will explore the end-to-end workflow of managing foundation model (FM) development on Amazon SageMaker HyperPod. Our discussion will cover both distributed model training and inference using frameworks like PyTorch and KubeRay. Additionally, we will dive into operational aspects, including system observability and resiliency features for scale and cost-performance using Amazon EKS on SageMaker HyperPod. By the end of this hands-on session, you will gain a robust understanding of training and deploying FMs efficiently on AWS. You will learn to leverage cutting-edge techniques and tools to ensure high performance, reliable, and scalable FM development.

Location: Room 206

Duration: 1 hour

Author:

Mark Vinciguerra

Assoc. WW Solution Architect

AWS

Mark Vinciguerra is an Associate Specialist Solutions Architect at Amazon Web Services (AWS) based in New York. He focuses on Generative AI training and inference, with the goal of helping customers architect, optimize, and scale their workloads across various AWS services. Prior to AWS, he went to Boston University and graduated with a degree in Computer Engineering.

Author:

Aravind Neelakantan

WW Solution Architect

AWS

Aravind Neelakantan is a Specialist Solutions Architect at Amazon Web Services (AWS), focusing on Generative AI training and inference solutions. His expertise spans artificial intelligence and high-performance computing, with research interests in performance analysis and optimization of AI workloads on heterogeneous architectures. Prior to AWS, Aravind developed reference architectures for large-scale AI clusters and specialized in AI inference performance optimization using hardware accelerators. He holds a Ph.D. from the University of Florida, specializing in heterogeneous HPC architecture.

Author:

Aman Shanbhag

WW Solution Architect

AWS

Aman Shanbhag is a Specialist Solutions Architect on the ML Frameworks team at Amazon Web Services (AWS), where he helps customers and partners with deploying ML training and inference solutions at scale. Before joining AWS, Aman graduated from Rice University with degrees in computer science, mathematics, and entrepreneurship.

Read more about Train and Deploy High-Performing AI Model Development at Scale

Anna Doherty

Partner

G9 Ventures

Runtime Attested HPC Cluster Reference Architecture for Confidential Computing

The rapid evolution of high-performance computing (HPC) clusters has been instrumental in driving transformative advancements in AI research and applications. These sophisticated systems enable the processing of complex datasets and support groundbreaking innovation. However, as their adoption grows, so do the critical security challenges they face, particularly when handling sensitive data in multi-tenant environments where diverse users and workloads coexist. Organizations are increasingly turning to Confidential Computing as a framework to protect AI workloads, emphasizing the need for robust HPC architectures that incorporate runtime attestation capabilities to ensure trust and integrity.

In this session, we present an advanced HPC cluster architecture designed to address these challenges, focusing on how runtime attestation of critical components – such as the kernel, Trusted Execution Environments (TEEs), and eBPF layers – can effectively fortify HPC clusters for AI applications operating across disjoint tenants. This architecture leverages cutting-edge security practices, enabling real-time verification and anomaly detection without compromising the performance essential to HPC systems.

Through use cases and examples, we will illustrate how runtime attestation integrates seamlessly into HPC environments, offering a scalable and efficient solution for securing AI workloads. Participants will leave this session equipped with a deeper understanding of how to leverage runtime attestation and Confidential Computing principles to build secure, reliable, and high-performing HPC clusters tailored for AI innovations.

Location: Room 201

Duration: 1 hour

Author:

Jason Rogers

CEO

Invary

Jason Rogers is the Chief Executive Officer of Invary, a cybersecurity company that ensures the security and confidentiality of critical systems by verifying their Runtime Integrity. Leveraging NSA-licensed technology, Invary detects hidden threats and reinforces confidence in an existing security posture. Previously, Jason served as the Vice President of Platform at Matterport, successfully launched a consumer-facing IoT platform for Lowe's, and developed numerous IoT and network security software products for Motorola.

Author:

Ofir Azoulay-Rozanes

Director of Product Management

Anjuna

Ofir Azoulay-Rozanes is the Director of Product Management at Anjuna Security. He brings over 30 years of experience in the software industry, including 15 years in cybersecurity. His career spans software engineering and product management leadership roles. Before joining Anjuna, Ofir led security products at Imperva and JFrog.

Read more about Runtime Attested HPC Cluster Reference Architecture for Confidential Computing

Improve Price Performance for LLM Serving with vLLM on TPU & GKE

Dive into a hands-on workshop designed exclusively for AI developers. Learn to leverage the power of Google Cloud TPUs, the custom accelerators behind Google Gemini, for highly efficient LLM inference using vLLM. In this workshop, you will build and deploy Gemma 3 27B on Trillium TPUs with vLLM and Google Kubernetes Engine (GKE). Explore advanced tooling like Dynamic Workload Scheduler (DWS) for TPU provisioning, Google Cloud Storage (GCS) for model checkpoints, and essential observability and monitoring solutions

Location: Room 207

Duration: 1 hour

Author:

Niranjan Hira

Senior Product Manager

Google Cloud

As a Product Manager in our AI Infrastructure team, Hira looks out for how Google Cloud offerings can help customers and partners build more helpful AI experiences for users. With over 30 years of experience building applications and products across multiple industries, he likes to hog the whiteboard and tell developer tales.

Author:

Don McCasland

Developer Advocate Lead

Google Cloud

Don leads the Cloud Developer Relations team for AI Infrastructure at Google Cloud. A 20-year veteran in Developer Operations, he is focused on empowering the global developer community to build and scale the next generation of AI applications on Google's cutting-edge platforms.

Read more about Improve Price Performance for LLM Serving with vLLM on TPU & GKE

Julianne Kur

Principal

Alliance Consumer Growth

The Future of Enterprise AI Infrastructure: What The AI Deployments of Today Will Mean for Enterprise Deployments in the Future

DataBank, one of the nation’s leading data center operators, with more facilities in more markets than any other provider, has seen the future of enterprise AI infrastructure and knows how to help enterprises get there.

With a customer base that spans 2500+ enterprises – in addition to hyperscalers and emerging AI service providers – DataBank has a unique perspective on the trends and lessons learned from customer AI deployments to date, which include some of the industry’s first NVL72/GB200 installations.

In this 60-minute session, John Solensky, DataBank’s VP of Sales Engineering, and Mike Alvaro, DataBank’s Principal Solutions Architect, will share what DataBank has learned from its early GPU installations for hyperscalers and AI service providers, how those lessons were applied to later enterprise installations, the impact that next-generation GPUs are having on data center designs and solution costs, and the lessons for future enterprise deployments.

Location: Room 206

Duration: 1 hour

Author:

Mike Alvaro

Principal Solutions Architect

DataBank

Michael Alvaro brings over 12 years of industry expertise spanning construction and mathematics to his role as Principal Solution Architect at DataBank, where he serves as technical lead for the data center sales team. Specializing in enterprise colocation solutions, Michael guides organizations through complex infrastructure requirements involving high-performance computing deployments.

As AI workloads rapidly scale across enterprises, Michael has become a trusted advisor for deployments demanding both high-density air cooling and advanced liquid cooling solutions. His unique construction background provides critical insight into physical infrastructure challenges, while his mathematical foundation enables precise optimization of power, cooling, and space efficiency. Michael’s approach centers on translating complex technical requirements into actionable deployment strategies, helping clients understand not just what’s possible, but what’s most cost-effective and operationally efficient.

Author:

John Solensky

VP of Sales Engineering

DataBank

John Solensky is the Vice President of Solutions Engineering at DataBank, where he leads a team focused on delivering colocation, cloud, and AI infrastructure solutions. With over 26 years of industry experience , John brings deep expertise in helping enterprises design and deploy secure, scalable, and high-performance platforms.

Since joining DataBank in 2020, John has been instrumental in advancing the company’s solutions engineering strategy, enabling customers to modernize IT environments and harness the power of AI-driven applications hosted in DataBank’s facilities. His leadership emphasizes collaboration, technical excellence, and a client-first approach, ensuring that organizations can rely on DataBank for mission-critical workloads and next-generation innovations.

Author:

Greg McNutt

Technical Director

Pure Storage

Greg McNutt is the Technical Director at PureStorage, Inc. where he has spent nearly a decade developing efficient methods of utilizing expensive hardware and limited power and supporting relatively high touch engineering labs. Over his career has has spent many years working on products from lowest level facilities to cloud and high scale products. Greg studied dependable computing at Stanford University. https://www.linkedin.com/in/gcmcnutt

Read more about The Future of Enterprise AI Infrastructure: What The AI Deployments of Today Will Mean for Enterprise Deployments in the Future

From Rack to Response: Build & Deploy Generative AI in 30 Minutes with NeuReality

Experience the future of GenAI inference architecture with NeuReality’s fully integrated, enterprise-ready NR1® Inference Appliance. In this hands-on workshop, you'll go from cold start to live GenAI applications in under 30 minutes using our AI-CPU-powered system. The NR1® Chip – the world’s first AI-CPU purpose built for interference – pairs with any GPU or AI accelerator and optimizes any AI data workload. We’ll walk you through setup, deployment, and real-time inference using models like LLaMA, Mistral, and DeepSeek on our disaggregated architecture—built for smooth scalability, superior price/performance and near 100% GPU utilization (vs <50% with traditional CPU/NIC architecture). Join us to see how NeuReality eliminates infrastructure complexity and delivers enterprise-ready performance and ROI today.

Location: Room 201

Duration: 1 hour

Author:

Paul Piezzo

Enterprise Sales Director

NeuReality

Author:

Gaurav Shah

VP of Business Development

NeuReality

Author:

Naveh Grofi

Customer Success Engineer

NeuReality

Read more about From Rack to Response: Build & Deploy Generative AI in 30 Minutes with NeuReality

Scaling LLM Inference with vLLM and AWS Trainium

Join us in this hands-on workshop to learn how to deploy and optimize large language models (LLMs) for scalable inference at enterprise scale. Participants will learn to orchestrate distributed LLM serving with vLLM on Amazon EKS, enabling robust, flexible, and highly available deployments. The session demonstrates how to utilize AWS Trainium hardware within EKS to maximize throughput and cost efficiency, leveraging Kubernetes-native features for automated scaling, resource management, and seamless integration with AWS services.

Location: Room 206

Duration: 1 hour

Author:

Asheesh Goja

Principal GenAI Solutions Architect

AWS

Asheesh Goja is Principal Gen AI Solutions Architect at AWS. Prior to AWS, Asheesh worked at prominent organizations such as Cisco and UPS, where he spearheaded initiatives to accelerate the adoption of several emerging technologies. His expertise spans ideation, co-design, incubation, and venture product development. Asheesh holds a wide portfolio of hardware and software patents, including a real-time C++ DSL, IoT hardware devices, Computer Vision and Edge AI prototypes. As an active contributor to the emerging fields of Generative AI and Edge AI, Asheesh shares his knowledge and insights through tech blogs and as a speaker at various industry conferences and forums.

Author:

Pinak Panigrahi

Sr. Machine Learning Architect - Annapurna ML

AWS

Read more about Scaling LLM Inference with vLLM and AWS Trainium

GIGABYTE AI TOP: Train Your Own AI on Your Desk

GIGABYTE AI TOP is a groundbreaking desktop solution that empowers developers to train their own AI models locally. Featuring advanced memory offloading technology and support for open-source LLMs, LMMs, and other machine learning models, it delivers enterprise-grade performance in a compact desktop form factor. This solution enables both AI beginners and professionals to build, fine-tune, and deploy state-of-the-art models with enhanced privacy, flexibility, and security.

Author:

Charles Le

CTO, Channel AI Solutions

GIGABYTE

Dr. Charles Le currently serves as Chief Technology Officer of Channel AI Solutions at GIGABYTE. He leads the AI software division and is the architect behind GIGABYTE’s flagship platform, AI TOP Utility, which empowers developers and enterprises to train and deploy large AI models with ease.

He is an expert in the training, finetuning, and inference of LLMs, LMMs, and other machine learning models, with deep knowledge across algorithm design, hardware acceleration, and system integration.

Before joining GIGABYTE, Dr. Le spent four years applying deep learning to the development of radiative cooling materials for marine robotics. He also has six years of experience in structural health monitoring and modal identification for infrastructure under dynamic loads such as earthquakes and wind. More recently, he has applied AI to enhance business intelligence, hardware R&D, and service AI assistants using tools like LangChain and LLM deployment.

Read more about GIGABYTE AI TOP: Train Your Own AI on Your Desk

Normal EDA: AI-Native Verification Without the Rework

As specifications grow to hundreds of pages, traditional verification workflows struggle to maintain consistency, traceability, and speed. This session demos Normal EDA, which replaces subjective, hand-written flows with NormML - a proprietary formal language that ingests raw specs, timing diagrams, and existing testbenches to build an auditable graph that auto-generates zero-to-one test plans, SystemVerilog/UVM stimulus, and traceable coverage links. The system reasons across multimodal data to flag inconsistencies before RTL reaches the simulator, slashing coverage closure time.

Author:

Maxim Khomiakov

Senior AI Engineer

Normal Computing

Maxim Khomiakov, PhD, is a Senior AI Engineer at Normal Computing, building reliable, high‑performance AI software for automating chip verification and design. He previously developed and deployed production-scale machine learning models within Apple Maps. Before that, he led data science efforts at Otovo and co‑founded Sunmapper (acquired by Otovo). Maxim holds a PhD in Machine Learning from the Technical University of Denmark (DTU).

Read more about Normal EDA: AI-Native Verification Without the Rework