A clinical research nurse opens a patient's record at 9 a.m. to check whether one person might qualify for one cancer trial. There are scanned pathology reports, free-text oncology notes, imaging impressions, a decade of lab values. By the time she has an answer, it is afternoon. Multiply that by every patient in a health system and you have the reason most clinical trials are slow, expensive, and chronically short of participants. Dyania Health built an AI to do that read in roughly half a second.
The problem they sawMost of medicine's data can't be searched
Roughly 80% of clinical information lives in unstructured text - the things a doctor types, dictates, or scans rather than the tidy codes and lab numbers a database can query. That text is where the diagnosis, the staging, the nuance actually lives. It is also, conveniently, the part no algorithm could reliably read. So health systems did the obvious thing: they paid humans to read it, one chart at a time, forever.
The cost shows up at the worst possible moment - clinical trial recruitment. A trial needs patients who match a long, fussy list of eligibility criteria. Finding them means reading charts. Reading charts means time. Time means trials that drag, miss enrollment targets, or quietly fail to represent the patients they were meant to help.
The founders' betFrom $60B in deals to medical NLP
Eirini Schlosser did not come from medicine. She came from Morgan Stanley, where she worked on more than $60 billion in technology, pharma, and consumer transactions before deciding that the interesting problem was in the data, not the deals. She had already founded Chuz, an NLP-powered recommendation engine, when she started Dyania Health in 2020.
Her bet was specific: large language models tuned for medicine - not general-purpose chatbots - could read a clinical note the way a trained clinician does, and do it at machine speed. To make that credible, she paired NLP applied scientists with an in-house team of physicians who design the criteria, assess protocols, and check the work. The leadership bench pulled from Amazon Alexa AI, Bloomberg AI, Flatiron Health, and NYU Langone.
Dyania is from the Greek diania (διάνοια) - exceptional intelligence, acumen, genius. The product is called Synapsis. Schlosser splits the company between Jersey City and Athens, which makes the etymology less of a marketing flourish and more of a homecoming.
The productSynapsis AI, and the half-second read
Synapsis AI is a proprietary, medically trained large language model. Point it at an electronic medical record - physician notes, pathology reports, imaging notes, plus the structured labs and codes - and it deduces answers to complex clinical questions at roughly 95% accuracy, in under 0.5 seconds per record. A human doing the same review typically needs 30 minutes or more.
The crucial design choice is where it runs: locally, inside the health system's own firewall. Patient data never leaves the building, which is how Dyania keeps the whole thing HIPAA and GDPR compliant. From that one capability comes a product line - Patient Finder for trial pre-screening, automated chart abstraction for observational studies, study feasibility assessment, and registry reporting that would otherwise eat weeks of staff time.
Synapsis AI
The core medically trained LLM. Reads unstructured + structured records, answers complex clinical questions at ~95% accuracy, ~0.5s per chart.
Patient Finder
Turns records into searchable structured data and precision-matches patients to trial eligibility criteria across the whole system.
Chart Abstraction & Registries
Automates review for observational studies, feasibility, and complex registry reporting - the manual work nobody enjoys.
Three products, one trick: read the note faster and more consistently than a tired human at 4 p.m.
Dyania Health is founded
Eirini Schlosser sets out to automate manual medical chart review across Jersey City and Athens.
$10M Series A
Led by HealthX Ventures with Tech Square Ventures and Cleveland Clinic Ventures. Total raised reaches ~$17.55M.
CancerX accelerator
Selected for the public-private program aimed at accelerating innovation against cancer.
Cleveland Clinic goes enterprise-wide
After pilots in cardiology, oncology, and neurology, the clinic rolls Synapsis out across its research system.
The proofThe numbers, before you take their word for it
Claims about AI accuracy are cheap. Dyania's case rests on two trials run with Cleveland Clinic. In a melanoma trial, Synapsis identified eligible patients in 2.5 minutes at 96% accuracy. The human comparisons: a specialized nurse took 427 minutes at 95%, and an oncology nurse took 540 minutes at 88%. Faster and at least as accurate is the only version of this story that matters.
One melanoma patient. Three reviewers.
The chart that makes a research coordinator's eye twitch - in the good way.
The cardiology result is arguably the bigger one. In the DepleTTR-CM trial, Synapsis analyzed 1.2 million patient records and reviewed 1,476 of them in a single week. It surfaced 30 eligible participants. Standard recruitment had found 14 over 90 days. More patients, found faster, across more sites - which also means a trial population that looks more like the real one.
The missionUnlocking the data that was already there
Dyania's stated vision is to change how health systems and life sciences companies interact with the unstructured data trapped in clinical records. Put plainly: the insight was always in the chart. The job is to make it readable at scale, without shipping patient data anywhere or asking clinicians to spend their careers as human optical character recognition.
The company says its models consistently outperform human experts at chart review. That is a strong claim, and the honest caveat is that "outperform" depends on the task and the trial. But the direction is hard to argue with: a tireless reader that is fast, consistent, and stays inside the firewall is a genuinely useful colleague for research that has always been bottlenecked on attention.
Cleveland Clinic is both customer and investor through Cleveland Clinic Ventures, and may benefit from sales of the technology. Worth knowing when you read the trial numbers. The numbers are still the numbers.
Why it matters tomorrowBack to the nurse at 9 a.m.
Return to where we started. The nurse who once spent a morning on a single chart now gets a precision-matched shortlist before her coffee cools, and spends her day on the part of the job that needs a human: judgment, the patient conversation, the consent. The trial that used to crawl fills faster and fairer. The patient who would have been missed gets the call.
That is the whole bet, and it is a narrow one - which is why it might work. Dyania Health did not set out to reinvent medicine. It set out to read the chart. Do that fast enough, accurately enough, and quietly enough to satisfy a compliance officer, and the slowest step in clinical research stops being the slowest step. Forty-five people in Jersey City and Athens are betting the rest follows.