
13 Nov 2026
Dyania's hospital AI reads the chart on NVIDIA boxes that stay inside the firewall
NVIDIA and Dyania Health described Synapsis AI on Nov. 13, 2026, a clinical reasoning system that runs on a hospital's own NVIDIA servers so patient records never leave the building.
Hospitals have been sitting on decades of notes that no person can reread at that scale. Running the model on the hospital's own machines, so the records never leave the building, is the right instinct, and a 7,812-patient valve study is a real use, not a demo reel. The catch is the usual one: these are the companies' own accounts of what the system can do, and a doctor still has to check the answers.
Dyania Health, a clinical AI company in New Jersey, built Synapsis AI. It calls the product a clinical reasoning engine. A reasoning engine, here, means software that reads a chart and connects the findings, rather than only searching for a keyword. NVIDIA says it works across a patient's notes, operative reports, imaging and pathology reads, labs, and discharge summaries. It connects findings that may be years apart, and it returns clinical signals that are each traceable to their source in the record. A clinical signal is a finding the software thinks a clinician should see. Clinicians review the evidence and make every decision. NVIDIA describes Synapsis AI as an assistive tool, not a replacement for a physician's review. Those lines are NVIDIA's, in the Nov. 13, 2026 blog post.
Synapsis AI runs on NVIDIA-accelerated computers inside a health system's own firewall. A firewall, here, is the boundary that keeps a hospital's network off the public internet. NVIDIA says the entire system's patient records can be processed on site, and that patient data is never shared and never leaves the hospital. Those lines are NVIDIA's.
Dyania builds and trains its clinical models with NVIDIA's open training libraries, then runs them on the hospital's own GPU servers. A GPU is the chip that does the heavy math. NVIDIA names the training libraries NeMo, and the tools that run the models TensorRT and TensorRT-LLM. Those product names are NVIDIA's. The plain version is that the model is trained with NVIDIA's tools and then runs on machines the hospital keeps inside its walls.
Cleveland Clinic used Synapsis AI on more than 7,800 patients who had an aortic valve replacement between 2005 and 2021. Fierce Healthcare specifies 7,812. An aortic valve replacement is surgery that swaps the valve letting blood out of the heart. NVIDIA says the system reasoned across each record, including later echocardiograms, to help build a model that predicts valve deterioration and the chance a patient would need another procedure at the three-year, five-year, and ten-year marks. An echocardiogram is an ultrasound of the heart. NVIDIA says that chart review would have taken clinicians years to finish by hand. Fierce, citing Dyania founder and chief executive Eirini Schlosser, says the tool answered 88 questions per patient and linked each answer to the supporting lines in the chart. The count of more than 7,800, the years, and the "years by hand" line are NVIDIA's. The 7,812 and the 88 questions are Fierce's, citing Schlosser.
Dr. Lars Svensson, chief of the Heart, Vascular and Thoracic Institute at Cleveland Clinic, said: "Using technology to bring together and analyze information that can help us predict the duration of implanted heart valves is a promising step towards enhanced outcomes and more personalized care for patients." That quotation is his, in both NVIDIA's post and Fierce.
Fierce says Synapsis AI is built on NVIDIA's enterprise software from end to end. Schlosser said it does not need to "speak to the internet" or interact with a cloud model or a frontier model. A frontier model is one of the largest general models, the kind a vendor runs in its own cloud. Dyania says it has invested more than 25,000 physician hours annotating medical cases. Annotating means a doctor marked up real charts so the model had examples to learn from. The company launched out of stealth two years ago. Stealth means it was working before it told the public. Cleveland Clinic began working with Dyania about a year ago on clinical trial recruitment, the work of finding patients who fit a study. Those lines are Fierce's, from the interview with Schlosser.
Under the collaboration described today, the companies will work to bring Synapsis AI, on NVIDIA's machines, to health systems in the United States and internationally. They will publish joint case studies and integrate the system with NVIDIA's healthcare software. They are also exploring a move from batch processing, running a pile of charts after the fact, toward insights that arrive close to real time at the point of care, meaning while the clinician is with the patient. Those plans are the companies'. They describe work ahead, not a switch that every hospital can flip today.
Schlosser told Fierce: "Every patient's story is spread across years of notes, reports and results, and for the first time we can reason across all of it at once, the way a specialist does, for every patient in a health system rather than one chart at a time." That quotation is hers.
Chelsea Sumner, NVIDIA's U.S. healthcare AI startups ecosystem lead, told Fierce that health systems need AI they can trust, integrate, and operate efficiently, and that success should be measured by whether it helps teams find patients who need attention and spend more time delivering care. That is her account, as Fierce reported it.
The picture is NVIDIA's product graphic of a Synapsis AI timeline. It runs from a March 2019 operative note through a 2029 projection, with a dotted organ-shaped map at each stop, and the line "PATIENT DATA IS PROTECTED INSIDE YOUR HEALTHCARE SYSTEM." The credit is NVIDIA.
In plain terms, Dyania's Synapsis AI reads years of a patient's chart on NVIDIA machines that stay inside the hospital, and Cleveland Clinic used it on 7,812 aortic-valve records to help estimate how long an implanted valve might last. Each answer is supposed to point back at the chart. A clinician still has to check it. What the system can do is the companies' own account.
RELATED
- Mayo Clinic puts an AI clinical advisor inside the medical record
- Doctors preferred AI-written hospital discharge summaries 95% of the time in a Stony Brook study
- Orakl Oncology raises $34M to test cancer drugs on "virtual patients" first
- Mosaic comes out of stealth with $47M to teach catheters to find their own way through blood vessels
Sources
- NVIDIA — care teams and Synapsis AI, 13 Nov 2026
blogs.nvidia.com
- Fierce Healthcare — Dyania, NVIDIA, Cleveland Clinic, 13 Nov 2026
fiercehealthcare.com