The company mining hospitals' untapped clinical data - notes, images, and raw ECG waveforms - to make drug development faster and healthcare AI more honest.
Walk into almost any hospital and you will find the most detailed record of human health ever assembled - and almost none of it reaches the people trying to build better drugs and safer algorithms. Electronic records capture diagnoses. Clinicians dictate long, unstructured notes. Machines produce ECG waveforms, echocardiogram videos, and imaging scans by the terabyte. Most of it is locked away, siloed inside individual health systems and formatted in ways no research team can easily use.
Dandelion Health, a New York company, was built to unlock that data responsibly. It partners with hospital systems to gather de-identified, multimodal patient records, then turns them into research-ready datasets and software that pharmaceutical companies and AI developers can actually work with. The scale is substantial: roughly 73 hospitals, more than 15 million patients, and over 8 petabytes of structured, unstructured, and raw biological data.
What distinguishes the approach is breadth. Rather than pulling from a single academic medical center - the usual source of clinical research data - Dandelion deliberately assembles a network spanning different geographies and demographics. Its early health-system partners included Sharp HealthCare in urban California, Sanford Health in rural South Dakota, and Texas Health Resources across the Southwest. The point is to reflect the full spread of American patients, not a narrow slice.
That design choice is not cosmetic. Healthcare AI has a well-documented tendency to work in the lab and stumble in the world, in part because models are trained and tested on populations that do not resemble the patients they later encounter. Dandelion's bet is that better, more representative data - not just bigger models - is what makes clinical AI trustworthy.
"We built Dandelion to change how untapped clinical data gets utilized in drug development."
- Elliott Green, CEO & Co-FounderDandelion's platform stretches across the pharmaceutical lifecycle - from designing a trial before it runs, to validating an algorithm after it ships.
The multimodal foundation: structured records and claims combined with clinical notes, medical images, and raw biological signals such as ECG waveforms and echocardiograms.
Core platformApplies vetted AI algorithms to turn unstructured clinical text into structured, research-ready information - and to validate models against real-world populations.
Launched 2024Uses real-world patient simulations to optimize clinical trial design, sizing, and eligibility criteria before a single patient is enrolled.
2026Enables disease-trajectory modeling, biomarker discovery, and real-world evidence generation across large longitudinal cohorts.
2026Validates AI algorithms and digital biomarkers against diverse real-world patient populations they were never trained on.
2026The first truly multimodal real-world library for the GLP-1 drug class - full longitudinal records with roughly 200,000 patients on GLP-1 agonists.
Launched 2024A clinical trial can consume hundreds of millions of dollars and years of runway. In one published proof point, Dandelion emulated a trial for a top-10 pharmaceutical company using real-world data - and pointed to a materially cheaper, faster path.
Plenty of companies sell real-world data. Dandelion's differentiators are the combination of modality and representation: it does not stop at tidy spreadsheet records but ingests notes, images, and raw waveforms, and it sources across rural, urban, and suburban systems rather than a single hospital's population.
The business is business-to-business. Dandelion licenses access to its de-identified dataset and sells software and services to two audiences: life-sciences companies that need trial design, real-world evidence, and biomarker discovery, and AI developers who need to train, validate, and win regulatory support for their models. Data is gathered through partnerships with hospital systems - an arrangement that gives providers a stake in responsible reuse of their patients' de-identified records.
It sits in a competitive real-world-data market alongside players such as Truveta, Verana Health, Aetion, Komodo Health, Tempus, and Flatiron Health. Dandelion's wedge is multimodal depth and validation - not just how many patients are in the database, but how much you can actually learn from each one, and whether an algorithm holds up on patients outside its training set.
Dandelion's founding team is unusually cross-disciplinary - a mix of healthcare operating experience, clinical research, and behavioral economics pointed at the same problem.
Started in finance, then moved into healthcare at Oscar Health before founding Dandelion.
Physician-scientist at Harvard Medical School and UC Berkeley, known for research on bias in medical algorithms.
MIT economist and MacArthur Fellow whose work bridges machine learning and human decision-making.
Established in New York to unlock untapped, real-world clinical data for drug development and AI. (Founding year approximate.)
Raises roughly $15.9M from Primary Venture Partners, Moxxie Ventures, and Convergent Ventures.
Debuts the first multimodal real-world data library for the GLP-1 drug class, drawing on Sharp HealthCare, Sanford Health, and Texas Health Resources.
Raises a Series A led by Healthier Capital, with Colle Capital and existing investors, to scale its multimodal platform and three lifecycle products.
"The richest source of real-world patient data, paired with proprietary AI tools - so drugs and algorithms are tested on patients who look like the real world."
Dandelion serves global pharmaceutical and life-sciences companies, healthcare AI developers, and academic researchers. Publicly referenced work includes engagements with top-10 pharma companies, collaborations with academic medical-center researchers - including a team that built an algorithm to quantify fat, muscle, and bone changes from abdominal CT scans - and partnerships spanning the American Heart Association and a consortium of health systems.
Pharma teams use the platform to design smarter trials, generate real-world evidence, and expand drug indications.
Model builders train, validate, and pursue regulatory support against diverse, multimodal real-world data.
Academic and clinical researchers surface biomarkers and study disease progression at population scale.
It provides a multimodal real-world data platform and AI software that help pharmaceutical companies and AI developers design trials, generate real-world evidence, discover biomarkers, and validate algorithms using de-identified patient data.
Elliott Green (CEO), physician-scientist Ziad Obermeyer, and economist Sendhil Mullainathan, a MacArthur Fellow.
About $29.9M in total, including a $14M Series A led by Healthier Capital announced in May 2026.
It spans roughly 73 hospitals, more than 15 million patients, and over 8 petabytes of structured, unstructured, and raw biological data.
Global pharmaceutical and life-sciences companies, healthcare AI developers, and academic researchers.
Profile compiled from public sources. Figures such as founding year and funding breakdowns are approximate where noted.