The Doctor's AI Colleague: Kanza Deploys a New Standard for Clinical AI

SILICON VALLEY, CA – June 08, 2026 – In a quiet but significant deployment, the first instance of a new class of medical artificial intelligence has gone live in a US clinic. Kanza AI's Clinical Reasoning System (CRS) is now operational at Freyja Clinic in Redwood City, California, marking a pivotal moment in the evolution of digital health. This is not another AI scribe or a glorified search engine; it is a system designed to reason through complex medical cases alongside physicians, powered by a sophisticated infrastructure from Lightning AI and NVIDIA.

The launch represents a potential solution to one of the most persistent challenges in technology: how to deploy powerful AI in high-stakes, highly regulated environments without compromising on privacy, security, or performance. For healthcare, an industry grappling with data overload and physician burnout, the promise of a true AI partner has been a long-held, but largely unmet, ambition.

A New Kind of Clinical Colleague

The central distinction of the Kanza AI system is its mode of interaction. Unlike tools designed to passively document encounters or fetch information, the CRS actively participates in the diagnostic process. It weighs evidence, flags uncertainty, and, crucially, cites every step of its reasoning, making its thought process fully auditable and reproducible. This transparency is the bedrock of clinical trust.

This collaborative approach is precisely what makes it different. Dr. Jan Rydfors, a Stanford-trained OB-GYN who uses the system at Freyja Clinic, captures the sentiment perfectly. “It was designed for me as a physician and for my clinic. It reasons the way I do, to support me. A colleague, not a co-pilot or just a scribe,” he stated. “This is how medicine should work with AI.”

This sentiment is no accident. The system has been co-developed with a peer-nominated group of more than 30 physicians across leading institutions in the US and Japan. This clinician-led development ensures the technology is not merely imposed upon the workflow but is woven into its very fabric, addressing real-world needs and respecting the nuances of medical practice. By functioning as a reasoning partner, the CRS aims to augment, not replace, the physician's expertise, improving diagnostic accuracy and freeing up cognitive bandwidth to focus on patient care.

The Defensible Moat: Data and Proven Performance

At the heart of Kanza AI's breakthrough is a strategic asset that is nearly impossible for general-purpose AI companies to replicate: its data. The system is built not on publicly scraped web data, but on a proprietary substrate of over 300 terabytes of longitudinal clinical data. This massive, curated dataset, drawn from a network of over 90 hospitals and 400 clinic locations, provides the deep, real-world context necessary for nuanced clinical reasoning.

This “data moat” is what allows Kanza to fine-tune best-in-class open-source models into highly specialized clinical engines. The results are striking. In an independent evaluation of 1,060 randomly sampled cases, the CRS placed first on all 32 dimensions of clinical reasoning when compared against leading frontier models like GPT, Claude, and Gemini. On industry benchmarks like MedXpertQA and the USMLE medical licensing exam, it not only posts the highest accuracy scores but also demonstrates a significantly wider reasoning lead. Its reasoning layer lifts the performance of any model it runs on, demonstrating that the value lies not just in the model, but in the entire substrate.

“The organizations that will define the next decade of enterprise AI are sitting on proprietary data that general-purpose infrastructure cannot touch,” said Saurabh Giri, Chief Product & Technology Officer at Lightning AI. He notes that Kanza’s defensibility comes from its unique combination of data access, a knowledge graph, and a reasoning layer, which Lightning AI’s platform then takes to production at scale.

The Engine Room: Secure, Sovereign AI Infrastructure

Superior models and data are only part of the equation. Deploying them within the labyrinthine compliance and security constraints of healthcare is the other, often harder, part. This is where the partnership with Lightning AI becomes critical. Kanza’s CRS runs on GraphN, Lightning AI’s agentic AI platform, which acts as the execution layer for the entire system.

GraphN orchestrates the complex web of models, tools, and agents that comprise the CRS. Its most crucial feature for healthcare is its deployment flexibility. It allows the CRS to run in any environment a healthcare institution’s compliance regime demands—from the public cloud to a fully sovereign, on-premise, or even air-gapped system completely disconnected from the internet. This capability directly addresses the core healthcare challenges of data sovereignty, patient privacy under HIPAA, and the fear of vendor lock-in. By enabling institutions to keep their most sensitive data within their own walls, it removes a major barrier to AI adoption.

“Healthcare has a knowledge access problem—how to turn proprietary data into intelligence that works at the point of care,” explained Samir Arora, Founder and CEO of Kanza AI. “GraphN, running on Lightning AI’s NVIDIA infrastructure, provides the production backbone that helps deliver Kanza’s CRS.” This secure backbone ensures that as the system is used, every clinical decision helps it become more grounded, auditable, and localized to the institution it serves.

The underlying NVIDIA compute infrastructure provides the raw power necessary for these complex AI workloads, completing a hardware and software stack designed from the ground up for the demands of enterprise AI. This architecture demonstrates that for AI to succeed in high-stakes fields, the infrastructure must be as thoughtfully designed as the algorithm itself. It must be secure, compliant, and powerful enough to deliver intelligence at the point of care, without compromise.