In the Lyuliang Mountains on the Loess Plateau, one general practitioner serves more than a thousand villagers. He weighs each patient's history and medications, makes calls across dozens of conditions, and does it all before lunch. Like most rural doctors in China, he was never going to get a specialist down the road. So the road came to him as software.
The system on his desk is the Smart Medical Assistant (智医助理) from Xunfei Healthcare (讯飞医疗), a company listed in Hong Kong under stock code 02506.HK. It is one of the clearest examples of AI moving from hospital showcase to everyday clinic — not to replace the doctor, but to sit beside one.
What the assistant actually does
The tool draws on a knowledge base built from over 13,000 clinical guidelines, 1,400 standard clinical pathways, and a database covering more than 6,000 diseases (per Xinhua / the State Council Information Office's WAIC 2026 reporting). When a patient comes in, it suggests possible diagnoses and treatment options, checks new prescriptions against the patient's existing medications, and highlights symptoms that need further examination. It follows national treatment protocols rather than improvising.
Crucially, it does not close the case. The doctor decides. The assistant's job is to widen the room a single overworked clinician has to think in — to surface the thing that might otherwise be missed at the end of a long day.
How far it has reached
According to Xinhua's reporting from the 2026 World AI Conference, the system has been rolled out to 811 county-level regions nationwide, supporting more than 250,000 doctors in primary health care institutions, and has delivered over 1.2 billion diagnostic recommendations. Xunfei Healthcare states the assistant was the world's first AI system to pass China's national medical licensing exam (written test), a claim the company lists on its own corporate page.
The company is also piloting AI-based chronic disease management for more than a dozen conditions, including diabetes, nudging the system from treatment toward prevention.
From clinic to life record
The same underlying model is being used to build resident health portraits (居民健康画像) in Shanghai. Across districts including Jing'an (静安), Xuhui (徐汇), Changning (长宁) and Minhang (闵行), the system has compiled dynamic, whole-life-cycle health files for roughly 1.49 million contracted residents, according to Jiefang Daily's coverage of the WAIC showcase. The payoff is mundane but large: a doctor's lookup of a patient's full history dropped from about five minutes to roughly thirty seconds.
That matters because the bottleneck in primary care is rarely the absence of data — it is the time to read it. A portrait that a clinician can scan in half a minute changes how many patients a morning can safely hold.
Why this is the interesting kind of deployment
There is a steady drumbeat of "AI passed another medical exam" stories. This one is different in two ways. First, the scale is real: hundreds of thousands of doctors and over a billion suggestions is not a demo. Second, the design accepts a hard limit — the assistant informs, the physician decides — which is the only posture that survives contact with actual patients.
Xunfei Healthcare is not alone, and the government has signaled intent to push further: national guidance on "AI + health" calls for intelligent assistance in primary care to be essentially covered by 2030.
What readers can do now
- If you are in China and see "智医助理" at a community health center running the assistant, treat its suggestions as a second opinion the doctor is already using — ask what it flagged.
- Patients with chronic conditions should ask whether their district runs an AI chronic-disease management program; diabetes pilots are among the earliest and may already be available locally.
- For health-system watchers elsewhere, track the "assist, don't decide" model as the template most likely to clear regulatory and trust barriers in primary care.
Honest limitations
The headline figures come from Xinhua and from Xunfei Healthcare's own disclosures, including its corporate profile and mid-year business reporting; the company is the operator, so adoption and recommendation counts are self-reported and not independently audited here. We did not locate a peer-reviewed study in this pass measuring diagnostic-error reduction attributable to the assistant, nor data on false negatives — cases where the AI stayed silent and the doctor also missed something. County coverage (811 regions) reflects a July 2026 snapshot and grows over time. The "passed the licensing exam" claim is the company's and has not been re-verified against the exam administrator in this article.
Key takeaways
- Xunfei Healthcare's Smart Medical Assistant supports 250,000+ primary-care doctors across 811 county-level regions, with 1.2 billion+ recommendations delivered.
- It was the first AI system to pass China's national medical licensing exam and is positioned as a decision aid, not a replacement.
- The bigger win is time: in Shanghai, AI-built health portraits cut a doctor's patient-history lookup from five minutes to thirty seconds for 1.49 million residents.