Tsinghua AIR-backed Zijing Zhikang upgrades AI hospital with multi-agent avatars

The company, founded two years ago and incubated by AIR, which was founded by Academician Zhang Yaqin, says each doctor can have a dedicated AI avatar. A single avatar can support a doctor when a case falls within one specialty. When a patient’s condition spans departments, avatars for different specialties can join the same consultation, giving the attending doctor broader specialist input without first assembling a traditional multidisciplinary team.

The report said the approach targets a longstanding bottleneck. Patients whose symptoms cross specialties often move from one department to another before a formal multidisciplinary consultation (MDT) is arranged, which can delay treatment and add cost. Zijing Zhikang’s answer is to bring specialist avatars into earlier stages, including pre-consultation and first visits, rather than waiting for a full MDT.

The system has two modes, according to the report. In an on-demand reminder mode, avatars intervene when new patient information suggests another specialty should be involved, without interrupting the normal workflow. In consultation mode, multiple avatars participate directly in cases that require joint analysis, presenting suggestions on diagnosis and medication for the doctor to compare.

Zijing Zhikang does not force the avatars to converge on one answer. The system displays each avatar’s recommendation, highlights disagreements and shows which avatar made which suggestion. The doctor then combines the patient’s condition, supporting evidence and clinical experience to make the final call. The report quoted an internal medicine physician as saying that AI is useful when it points to directions the doctor may not have considered, and that disagreements require careful review rather than automatic acceptance.

The avatars are designed to learn from real doctors. The technical route is based on SEAL, or simulacrum-based evolvable agent learning, proposed by Professor Liu Yang’s team at AIR when the Agent Hospital project began. The system uses a two-layer architecture: training in a virtual hospital where AI patients, doctors and nurses simulate high-volume care, followed by validation and fine-tuning in real-world practice.

In the virtual environment, successful cases enter a case library and failures are reviewed to extract reusable rules, according to the report. The company says this allows AI doctors to run through hundreds of thousands or even millions of cases in days, including rare diseases and complex combinations, at lower compute cost than training a large model from scratch. In real clinics, doctors’ feedback is fed back into the system so the avatars can adapt to practical constraints, such as cost and insurance coverage.

Zijing AI Hospital serves patients, doctors and hospitals. The patient side handles pre-consultation, health records and follow-up management; the doctor side provides avatars, medical records and decision support; the hospital side supports scheduling, permissions and data management. As of July 2026, the company said it had built 54 AI doctors and millions of simulated AI patients covering thousands of diseases. At Beijing Tsinghua Changgung Hospital, 18 clinical departments have begun outpatient testing, with more than 40 doctors using the system regularly and the company reporting a 30% gain in overall doctor efficiency.

The company also said its AI doctors scored above 96% on the MedQA dataset and achieved top results in several international medical evaluation datasets. It positions AI as an auxiliary tool and says it uses value alignment to turn medical-humanistic principles into constraints and preferences that the system can understand and execute, leaving final decisions to human doctors.

The company says the next step is to expand the system to more hospitals and bring in more specialty tools and partners.