AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

NVIDIA's Rama Akkiraju on Building the Right AI Infrastructure for Enterprise Success

· NVIDIA AI Podcast

In Episode 255 of the NVIDIA AI Podcast, host Noah Kravitz speaks with Rama Akkiraju, VP of IT for AI and ML at NVIDIA and former IBM Fellow who worked on Watson, about the emerging role of the AI platform architect and what it takes to operationalize AI inside large enterprises. Akkiraju frames her team's mission as driving AI adoption across NVIDIA itself, building chatbots, copilots, AI agents, and an enterprise-wide generative AI platform that other business units consume via APIs and low-code/no-code interfaces. She traces the evolution of enterprise AI from perception/classic ML through the generative AI era and into the current agentic AI phase, noting that the leap from generative to agentic took roughly two years versus the 25–30 years from perception to generative AI. She argues the impact on the enterprise is not just incremental automation but a fundamental rethinking of business processes and software—suggesting even that "SaaS may be dead" as AI logic moves into the business layers themselves. She walks through the complexity of building enterprise AI: data ingestion pipelines, RAG over vector databases, role-based access control, content security, model selection, evaluation, observability, and continuous improvement via data flywheels. Akiraju lays out the components of a full enterprise AI stack—enterprise data management, content security, vector databases and retrievers, compute/infra, GPU-optimized inference serving (mentioning NVIDIA NIM), LLM gateways, LLM-level observability, auto-evaluation frameworks, agentic frameworks (LangChain, LangGraph, and NVIDIA's own AgentIQ/AIQ released at GTC), low-code/no-code tooling, and fine-tuning/data flywheel pipelines. Looking ahead, she highlights three trends: unification of AI and traditional enterprise architecture, rise of domain-specific and smaller models alongside specialized hardware (including edge/mobile/browser), and the proliferation of more autonomous agentic systems requiring long-term memory, context management, and workflow chaining—with NVIDIA's Dynamo and related hyperscale compute stack positioned as the infrastructure response. She points listeners to ai.nvidia.com for workflows, blueprints, and code samples.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.