ADELPHI, Maryland—In a quiet voice and in her native Spanish, the woman explained to Dr. Shantanu Nundy that she had been feeling dizzy whenever she stood up.

She cleaned houses and worked in a store. There was a lot going on at home—and now this. She choked up describing it all.

Nundy’s clinic, called Mary’s Center, is a primary-care practice, and hers was a classic primary-care problem: common, yet strange; vague, yet worrisome—troubling enough to send the woman to the emergency room the day before, sticking her with a $200 bill. Still, the dizzy spells were not definitive enough for the ER to do anything about.

Nundy suspected she had something deep inside her ear that was throwing off her balance. To make sure, he had her perform something called the Dix-Hallpike test: From a sitting position, he asked her to fall back onto the exam table, then toss her head to one side. That would help determine whether the source of the dizziness was a problem in the inner ear.

It didn’t really work. When she sat back up, she felt fine.

Nundy stepped into the hallway and wrote up her case in the clinic’s electronic medical record. But he still wanted to be sure the cause of the dizziness wasn’t a small stroke or something more serious.

He opened a new tab on his computer and went to a new website that he helps design and run: the Human Diagnosis Project, or Human Dx. The project allows primary-care doctors to ask for assistance on difficult cases from an online network of physicians all over the world.

He clicked “get help on a case” and, on a checklist-style page, input that she was “43f”—a 43-year-old female—with episodic dizziness for the past two months. He then submitted the case to a doctor at another Mary’s Center clinic, as well as to Human Dx’s entire database of nearly 7,000 doctors.

Trained in internal medicine, Nundy now leads the nonprofit arm of Human Dx, but he spends Fridays at the clinic as its only provider for adults. (Other doctors and nurses see children there the other days of the week.)

Mary’s Center is a safety-net clinic, so its patients pay according to their income. At just after 8:30, the waiting room was bustling. The staff issued each patient a number and called them back in English and Spanish—“Twenty-six ... veintiseis!”

Nundy says about 80 percent of his patients are uninsured, in some cases because of their immigration status. Even for those with insurance, a specialist might be out of reach because of high deductibles and co-pays or long wait times.

“For you and me, someone who has insurance, the standard of care is that you see an expert who lives and breathes ... your diagnosis,” Nundy says. But for the 28 million uninsured Americans, seeing, say, a dermatologist or a neurologist usually means getting on long waiting lists for a doctor who is willing to volunteer his or her time.

Human Dx might help doctors confirm their suspected diagnoses or think of things to rule out. At Mary’s Center, one man came in complaining of headaches and nausea, and the Human Dx physicians suggested a blood test called an ESR. Another time, Nundy used it to confirm a suspected case of rheumatoid arthritis before putting a low-income patient on a heavy-duty course of medications.

Experienced doctors use Human Dx for their most difficult cases, and newer providers use it to hone their skills. Johns Hopkins Hospital and other teaching hospitals are now using it to train medical residents. Georgia Lewis, a nurse practitioner who works with Nundy, used Human Dx when, two months into her stint at Mary’s Center, all the other providers went on vacation. Rashes can be confounding, so she’ll upload them to Human Dx along with a photo.

The contributors to the project are vetted based on how accurately they’ve solved past cases. Human Dx uses machine learning, which means that eventually the algorithms powering the diagnosis suggestions will become “smarter” based on the input of the doctors using it. The hope is that, over time, Human Dx can help reduce misdiagnoses, which according to studies happen up to 20 percent of the time.

Human Dx hopes to soon roll out to all 1,300 safety-net clinics in the United States. Ron Yee, the chief medical officer of the National Association of Community Health Centers, is helping clinics like Mary’s Center start using the platform. “We thought we can really help our communities because we have challenges getting specialty care,” he said.

Yee and his colleagues are still figuring out how to fit Human Dx into so many primary-care doctors’ workflows. They’re also puzzling through that eternal health-care question: how to get paid for it. “Does insurance accept this?” Yee said. “I don’t know what it looks like.”

Nundy acknowledges that Human Dx adds time to a doctor’s day. But he says researching difficult cases already adds time, as does reading reference materials or calling his med-school friends for their advice. He hopes that as the project progresses, it could count toward doctors’ continuing medical education, licensing requirements, or student loans. Eventually, he hopes to get all the area’s specialists who treat the uninsured on Human Dx, so they can offer their counsel digitally and save their charity care for those who really need to be seen in person.

It usually takes about six hours to get a response through Human Dx, but a little over an hour after Nundy had seen the woman with the dizzy spells, a few responses had already trickled in. The relative likelihood of the doctors’ guesses were represented by little green bars, like a Wi-Fi signal. The most common suggested diagnosis was dehydration, followed by stress, a ministroke, or Ménière’s disease, a disorder of the inner ear. “Now when I see a person with dizziness,” Nundy said, “I’ll think about Ménière’s disease.”

Most likely, the woman was just stressed and tired. But for Nundy and other primary-care doctors using Human Dx, it’s worth carefully considering “the consequences of being wrong. If this was my mom or my sister ... that’s what we would want,” he said. “That’s what patients deserve.”