From trade-show demo to a real job at the triage desk

"Which department for stomach pain?" "How do I get to gastroenterology?" "Is Dr. Wang in clinic today?" At the outpatient triage desk of Peking University People's Hospital, these questions are now fielded not only by nurses but also by a "silicon triage assistant" — Zhiyuan's Lingxi X2 humanoid robot. According to multi-source reports on October 16 (Shangguan News, Wenhui Daily, huanqiu.com), the smart-triage project jointly delivered by Zhiyuan Robotics and Ant Group's Afu has gone live at the hospital, extending AI medical services from online apps into offline outpatient scenarios.

Unlike dancing demo humanoids, this is a formal post: Lingxi X2 is stationed at the outpatient guidance desk, connected to Afu App's AI medical assistant, built on Ant's medical large model plus the hospital's private knowledge base, and answers with human-like voice and gestures.

Project file (cross-verified)

· Deployment: outpatient triage desk, Peking University People's Hospital (top-tier hospital)

· Robot: Zhiyuan Lingxi X2 (half-size humanoid)

· Model stack: Afu App AI medical assistant + Ant medical LLM + hospital-private knowledge base

· Scope: department guidance, visit-flow consultation, insurance Q&A, in-hospital navigation

· Operations (hospital statistics, as of Sept 30): 66 working days on duty, 46,686 cumulative interactions, about 707 per day, 1,420 on the busiest day

Why triage desks get embodied AI first

The outpatient hall is the most repetitive, high-frequency and fault-tolerant zone in a hospital: at peak hours, nurses juggle department queries, registration guidance, wayfinding and process explanations — all highly repeated, all demanding instant response. Kiosks can look up information, but they do not proactively greet a confused elderly patient clutching a lab report. Lingxi X2's value lies exactly in "mobile interaction + human-like accompaniment" — complementary to fixed terminals rather than a replacement.

The knowledge loop matters even more: grounded in the hospital's private knowledge base, answers are constrained to in-hospital information and standardized, avoiding the free-form risk of general-purpose LLMs in a medical setting. Sixty-six days and 46,686 interactions also mark the first continuously operating, at-scale dataset for embodied AI inside a public top-tier outpatient hall.

The first step of AI entering hospitals is not diagnosing patients — it is catching the questions no human has time to keep answering.

Connecting the dots: from hospital triage to factory lines

Zoom out: in the same week, Shanghai set a target of 8,000 humanoid robots in factories by 2028 (see our companion story), and Leju robots completed 40-second-cycle feeding tasks on an auto-parts line. Different scenes, same logic — embodied AI wins business not by "acting human" but by "being on duty": a defined post, quantified metrics, replicability.

For enterprise buyers, the recipe — high-frequency queries + private knowledge base + scenario engineering — is familiar: it is the same path NineZenith's Tianxing agent platform takes in government and enterprise service windows, letting agents absorb the repetitive questions first and reserving humans for complex judgment.

Trend watch: as the first cohort of robots "on the payroll" begins reporting operational data, AI procurement in hospitals, service halls and bank branches will shift from "can it demo" to operational KPIs — on-duty rate, interaction volume, escalation rate. (Information synthesized from public reports by Shangguan News, Wenhui Daily, huanqiu.com and Sina Finance)