Shubham Katiyar

About Shubham Katiyar

I am an executive AI and platform leader with 16+ years at Amazon, specializing in turning frontier research into durable, global-scale platforms. My work sits at the intersection of model science, distributed systems, hardware co-design, and business economics—where technical decisions are irreversible, timelines are compressed, and trade-offs carry material financial and customer impact.I currently lead Global Runtime Infrastructure and Inference Research for Amazon’s Nova foundation models, with end-to-end ownership of the systems that power understanding, generation, and multimodal AI across AWS Bedrock and Amazon’s largest internal businesses, including Alexa+, Retail, Prime Video, Ads, and Amazon Q. My scope spans inference-aware model architecture, disaggregated and real-time inference, kernel and networking optimization, hardware diversification (NVIDIA and AWS Trainium), and global production operations serving trillions of tokens with sub-100ms latency and four-nines reliability. I routinely make high-stakes launch and architecture decisions under uncertainty—owning short-term inefficiency to preserve long-term platform advantage and customer trust.Earlier, I was asked by the Amazon S-Team to reimagine Alexa for the LLM era following the “ChatGPT moment.” Rather than incrementally refactor a decade-old, intent-based system, I led a first principles rebuild around an LLM-centric, event-driven architecture—while preserving the latency and reliability customers expect for simple commands. That work collapsed dozens of legacy services into a small core, enabled multi-turn reasoning and safe real-world execution, and became the foundation for Alexa+.Across roles, I lead talent-dense organizations of distinguished and principal engineers and applied scientists, design decision environments that scale beyond my personal bandwidth, and operate with direct exposure to SVP and CEO-level stakeholders. What motivates me most is building platforms that outlast any single model or release—systems that translate frontier AI into reliable, economical capabilities real customers can depend on for the next decade.

Experience

1. Director, Agentic Runtime & Inference Amazon Nova (Agi) Aws Bedrock

   Amazon

   Mar 2023 — Present · Seattle, WA, US

   Executive owner of agentic runtime and inference for Amazon’s Nova family of foundation models, defining the price–performance, latency, and reliability foundations that enable planning, tool use, and long-horizon execution across both AWS Bedrock customers and Amazon’s largest internal businesses. Lead end-to-end systems spanning model architecture, distributed and real-time inference, hardware co-design, and global production operations.• Operated with direct S-team and CEO-level exposure, owning financial trade-offs involving billions of dollars in infrastructure spend and setting explicit recovery milestones to preserve long-term platform advantage.• Owned inference and agentic runtime for Nova’s multimodal foundation models, enabling agentic understanding and generation across text, image, audio, video, and documents for Alexa+, Retail, Prime Video, Ads, Amazon Q, and hundreds of external Bedrock customers.• Pioneered disaggregated agentic inference stack that matches each stage of inference to optimal hardware (CPU, commodity GPU, high-end GPU, accelerators), delivering 40–60% cost reduction from monolith baselines with negligible latency impact at scale.• Invented inference-aware model architectures across dense, MoE, long-context, and multimodal models, co-designing attention mechanisms, sparsity, speculative decoding, and mixed precision, enabling Nova to launch as the first Amazon model family deployed across multiple accelerator platforms (Nvidia GPUs and AWS Tranium); unlocked capacity, regional expansion, and sustained cost leadership.

Education

• Birla Institute of Technology and Science, Pilani

  Bachelor of Engineering (Hons.), Computer Science

Skills

• High Availability Architecture
• C++
• Service Oriented Architecture Design
• Scalability
• Hadoop
• Big Data
• Software Design
• Algorithms
• Project Planning
• Agile Methodologies
• Software Development
• Scrum
• Amazon Web Services (Aws)
• Software Project Management
• Java
• Project Management
• Web Services
• Data Visualization
• High Availability
• Gwt
• Data Analysis
• C
• Linux
• Data Processing
• Cloud Computing
• Amazon S3
• Lean Software Development
• Agile Methodolgy
• Distributed Systems