Healthcare IT · Collective intelligence
The Human Diagnosis Project
An open platform where doctors solve one another’s cases, built to hand specialist expertise to clinics that cannot buy it. Fourteen years on, the published science has held up better than the rollout.
A pair of swollen hands
In January 2017 a single mother in her thirties walked into a clinic on the north side of Washington, DC. Her hands hurt so much she had stopped working as a housekeeper. Months earlier another doctor there had referred her to a specialist, but the visit would have cost a few hundred dollars out of pocket, so she never went.
The internist who saw her, Shantanu Nundy, thought it was rheumatoid arthritis. The standard drugs are toxic enough that he did not want to start them on a hunch. So he typed the case into a web form, attached a photograph of her inflamed hands and pressed send.
Within hours rheumatologists he had never met began answering. By the next day they had confirmed the diagnosis. “I wouldn’t have had the expertise or confidence to be able to do that on my own,” Nundy told WIRED.
The gap it is aimed at
Roughly 30 million uninsured Americans depend on community health centers and free clinics. Those clinics can treat a cough. They cannot conjure a cardiologist. In its MacArthur Foundation proposal, Human Dx put the figure at 91 percent of safety net clinics unable to get fast, reliable specialist access for uninsured patients.
The field’s workaround is the electronic consult: a primary care doctor writes up a question, a specialist answers in writing, nobody books an exam room. Komarneni estimated to WIRED that eConsults could absorb 35 to 40 percent of specialist visits. San Francisco General’s eReferral system cut the average wait for a first consult from 112 days to 49 in its first year.
Human Dx made one bet that separated it from every other eConsult company: the specialists would not be paid. They would volunteer, the way people write Wikipedia entries, and machine learning would knit their answers into one ranked differential. That bet is the source of both its reach and its trouble.
The founder
Jayanth Komarneni came to medicine sideways. He completed five degree programs in six years at Penn and Oxford, including a Wharton MBA and an MSc in global health science. He advised healthcare clients at McKinsey and Bain, helped launch a global alternative investment firm, then went through Y Combinator.
He started the Project in 2012 and recruited Nundy in 2015 to pilot it in a real safety net clinic. By August 2017 the platform had passed 6,000 doctors in 70 countries and 500 institutions.
The year of the big number
On August 10, 2017 Human Dx announced the Human Dx Alliance for the Underserved. The American Medical Association joined, with the American Board of Internal Medicine, the American Board of Medical Specialties, the Association of American Medical Colleges, the Association of Clinicians for the Underserved, the National Association of Community Health Centers and the Dartmouth Institute. The plan: the eConsult system in 8,000 safety net clinics, with 50,000 specialists, serving more than three million patients in five years.
That same month Human Dx was one of eight semifinalists, from nearly 1,900 applications, for a single $100 million MacArthur grant. It did not win.
Skeptics arrived immediately. Ethan Weiss, a UC San Francisco cardiologist, told Axios he worried the system would end up “spitting out garbage.” Alice Chen, then chief medical officer of the San Francisco Health Network, warned WIRED of a two-tier outcome: some patients see a specialist, others get a computed composite of specialist opinions.
What followed was lopsided. The research came, and it was good. The scale did not, at least not on the public record. Nundy left for Accolade in October 2019, and the charity’s reported revenue peaked in 2018 and then fell away.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Collective accuracy | Barnett and colleagues, JAMA Network Open, March 2019: 1,572 cases, 2,069 users, 2014 to 2016. A correct diagnosis in the top three rose from 62.5 percent for individuals to 85.6 percent for groups of nine. | Public record |
| Pooled accuracy | Kurvers and colleagues, PNAS, August 2023: 1,333 Human Dx cases, each rated by ten diagnosticians. Single diagnosticians were 46 percent accurate; pooling ten reached 76 percent. | Public record |
| Human and machine | Zöller and colleagues, PNAS, June 2025: 40,762 physician differentials from Human Dx plus five large language models across 2,133 vignettes. Mixed groups beat doctors alone, doctor groups, single models and model ensembles. Four Human Dx people are coauthors, two of them disclosing personal financial interests. | Public record |
| Real referrals | Khoong and colleagues, Medical Decision Making, April 2022: 35 real referral requests from one San Francisco safety net institution, answered by 177 physicians. Collective advice was appropriate 69 percent of the time against 45 percent for individuals, and potentially harmful 54 percent against 41 percent. | Public record |
| Assessing doctors | Chatterjee and colleagues, JAMA Network Open, January 2019: 11,023 case attempts by 1,738 US physicians, trainees and students, validating a diagnostic performance score. | Public record |
| Adoption at scale | The 2017 target was 8,000 clinics, 50,000 specialists and three million patients in five years. No published report was found stating how many clinics enrolled or how many consults were delivered. | Not found |
| Legal structure | The Project calls itself a tandem charity and public benefit corporation. SEC records show Human Dx, Ltd., a Delaware corporation formed in 2012, selling $2,803,634 of equity to 15 investors in 2013. The Android app names it as developer, and a 2019 JAMA Network Open disclosure says the Project licenses the software used in that study. | Claim and filings differ |
| Charity finances | Forms 990 for Human Diagnosis Project Ltd, EIN 81-0865879: revenue of $925,000 in 2016, $1,002,362 in 2018 and $56,693 in 2021, the last year with a filing. ProPublica notes it “is not listed in the IRS’s most recent list of tax exempt organizations.” | Public record |
| Federal support | No NIH or NSF award, no FDA record and no registered trial found under the company name as of September 22, 2026. | Not found |
| Still running | The Android app was updated April 3, 2026, lists Human Dx, Ltd., San Francisco, as developer, and the website answered normally on September 22, 2026. | Public record |
Read plainly: the core scientific claim is among the better supported ideas in digital health. Groups of doctors beat individual doctors, repeatedly, across three peer-reviewed analyses of this platform’s own data. The weak link is not the science, it is delivery. The one study using real referrals rather than teaching cases found the collective both more appropriate and more likely to suggest something potentially harmful, and the charity built to carry the program to 8,000 clinics reported $56,693 of revenue in its final filed year.
What to watch
- The cooperative compensation model Komarneni described to Forbes in March 2026, paying contributors by the value of what they contribute. Volunteer labor was the original bet; this reverses it.
- Whether the charity files a Form 990 again, and whether its IRS master file listing returns.
- Whether the harm signal in the 2022 Medical Decision Making paper is reproduced or engineered away.
- Whether Human Dx, Ltd. files a second Form D. The last one was in 2013.
In their words
“We need to find solutions that scale the capacity of existing doctors to serve more patients at the same or cheaper cost.”
Jay Komarneni, founder and chair, to Scientific American, 2017 · Independent
“With other online tools if you help one patient you help one patient.”
Jay Komarneni, to WIRED, August 2017 · Independent
“Human Dx itself is organized as a tandem 501(c)(3) non-profit and public benefit corporation to enable a true partnership of the social, private, and public sectors.”
The Human Diagnosis Project, Alliance announcement on Medium, August 2017 · Company
“I’m not sure how you’d begin to demonstrate that it works or doesn’t work.”
Ethan Weiss, cardiologist, UC San Francisco, quoted by Fierce Healthcare, August 2017 · Independent
“It makes intuitive sense that [crowdsourced advice] would be better advice”
Ateev Mehrotra, Harvard Medical School, to Scientific American, 2017 · Independent
“You just have to find the sweet spot where you can leverage expertise and experience beyond traditional channels and at the same time ensure quality care.”
Alice Chen, chief medical officer, San Francisco Health Network, to WIRED, August 2017 · Independent
“while the doctors quoted had general options on technology in medicine, few were familiar with Human Dx and how it works”
Justin Hamilton, spokesperson, emailed to Fierce Healthcare, August 2017, quoted as published · Company
Related companies
Sources
- Public recordForm D, Human Dx, Ltd., CIK 0001588134
- Public recordForms 990, Human Diagnosis Project Ltd, EIN 81-0865879, FY2016 to FY2021
- Public recordComparative Accuracy of Diagnosis by Collective Intelligence
- Public recordAssessment of a Simulated Case-Based Measurement of Physician Diagnostic Performance
- Public recordComparison of Diagnostic Recommendations from Individual Physicians versus the Collective Intelligence of Multiple Physicians
- Public recordAutomating hybrid collective intelligence in open-ended medical diagnostics
- Public recordHuman-AI collectives most accurately diagnose clinical vignettes
- Public recordHuman Dx app listing, developer Human Dx, Ltd.
- IndependentWant a Diagnosis Tomorrow, Not Next Year? Turn to AI
- IndependentCan a Crowdsourced AI Medical Diagnosis App Outperform Your Doctor?
- IndependentDespite support from AMA, doctors worry the Human Diagnosis Project
- IndependentHow Collective Intelligence Could Soon Reshape Medical Decision-Making
- PartnerProject crowdsources specialists’ diagnoses for safety-net care
- CompanyA National Healthcare Alliance for the Underserved of America
- CompanySemifinalist profile and founder biography
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
