Published: 2026-07-07 · Tag: TCM · AI Medicine · Diagnostics

Beijing Start-Up Ships AI TCM Diagnostic Device to 12+ Languages at Global Digital Economy Conference

Guanwei Intelligent Technology (Beijing) Co., Ltd., a four-year-old start-up, displayed its facial-and-ocular TCM scanner at the 2026 Global Digital Economy Conference and confirmed the device is live in twelve language versions — English, Korean, Russian, Thai among them — with overseas clinics already piloting it.

It is, by some distance, the most internationally visible AI TCM diagnostic that has come out of China so far. Walk up to a booth, stand in front of the device, and the unit reads the face and the eyes for sixty seconds — by the time you stop blinking, the report is on the screen. The product manager calls it "the TCM examination a human practitioner would do, only faster and reproducible." Real-world use so far is small but steady, and the company is talking about expansion into the Middle East, Southeast Asia, and — of course — the global TCM-curious traveller who lands in Beijing and wants the scan.

What the device does, in plain language

A TCM practitioner doing a first consultation spends fifteen to thirty minutes on what is called the four diagnoses — looking, listening, asking, and feeling for the pulse (望闻问切). Most of the diagnostic value sits in the looking part. Skin colour on the cheeks, the tongue, the whites of the eyes, the way the person carries their shoulders — these are signals a trained practitioner reads against the nine-constitution model (九种体质) of TCM.

Guanwei's device compresses the looking step. A camera reads the face and the eyes, a vision model maps the visible features to the nine-constitution categories and to a small set of organ-system markers, and a report comes back describing which of the nine the user sits closest to plus a list of common symptoms associated with that type. The pulse-taking and the in-person conversation are still done by a human practitioner at a real consultation; the device is positioned as a front-end triage tool, not a replacement for the doctor. Beijing Daily's technology columnist described it as "a screener that knows how to wait politely for the doctor to finish."

The implications for an English-speaking visitor are practical. At a TCM hospital or wellness clinic the patient usually does not know what to ask for and does not know the right Chinese terms (痰湿体质? 阴虚?). The device closes the language gap — the report is in English, the recommendation list is in English, and the patient hands a printout to the practitioner when they sit down.

Where it showed up — and what Beijing is signalling

The Global Digital Economy Conference ran from 02 to 05 July 2026 at the China National Convention Center in Beijing. The conference was hosted by the Beijing municipal government and the National Development and Reform Commission; it drew 12,000 attendees and several hundred exhibitors. The Ministry of Industry and Information Technology, the Cyberspace Administration, and the Beijing municipal AI office were visible at the opening. Sponsorship came from Huawei, ByteDance, Xiaomi, and a long tail of Beijing-based start-ups.

Zhou Chao, the company's business manager, was on the floor with the device. His pitch was short: "We are using AI to digitise, standardise and modernise TCM diagnosis and treatment — to enable time-honoured medical wisdom to benefit more people worldwide." Behind him, a digital avatar on the conference welcome kiosk talked visitors through a simulated visit, which is how most attendees first met the technology.

"We are using AI to digitise, standardise and modernise TCM diagnosis and treatment, enabling time-honoured medical wisdom to benefit more people worldwide." — Zhou Chao, business manager, Guanwei Intelligent Technology (Beijing) Co., Ltd.

The conference programme for the first time included a dedicated AI-medicine track, jointly chaired by the Beijing Hospital Administration and the Tsinghua-IIIS joint lab. Three of the seven papers in that track were on TCM data infrastructure (舌象 database, constitution classification benchmarks, herbal-compound prediction models). Two of the three used Guanwei's data as the validation set. The writing is on the wall: the conference did not just put the device on display, it positioned TCM data as a strategic priority for the next stage of China's medical AI export.

The international side — twelve languages and counting

Zhou confirmed the device is shipping in twelve language versions as of July 2026: English (Simplified and Traditional), Korean, Japanese, Russian, Thai, Vietnamese, Indonesian, Arabic, French, German, Spanish, and Portuguese. Translation has been harder than the engineering. Each constitution type is grounded in TCM theory (痰湿, 阴虚, 气虚 etc.), and the team needed to find non-Chinese-language clinical terms that convey the underlying pattern without imposing a Western-medical redefinition. Korean and Japanese versions led the rollout because TCM has had clinical practitioners in those markets for decades and a pre-existing vocabulary.

Three overseas markets are in early pilot. A TCM chain in Seoul has placed two units in its flagship clinics, with plans to add four more by Q4 2026. A Moscow partner has placed one unit in an integrative-medicine practice, with plans for a Chinese-practitioner-staffed interpretation service. A Bangkok wellness centre has placed one unit. The company has not disclosed revenue or unit count, but said the manufacturing cost has come down 40% since the first-generation prototype in 2023, with the v3 model (the version on display) running on a Qualcomm Snapdragon-based edge-AI chip that does the inference on-device.

The Arabic version is the one to watch. The Gulf states have been buyers of TCM hospital services — Emirati patients have travelled to Beijing and Shanghai for TCM rheumatology consultations for at least five years. Putting a Mandarin-portable pre-screener in a Dubai clinic that already sends patients to Beijing changes the patient pipeline in interesting ways: the same constitution report the patient got in Dubai can be sent ahead to the receiving hospital, and the Chinese practitioner picks up the consult without a translation phone call.

What the device does NOT do

Three things to be clear about.

The device is not a clinical diagnostic. A TCM constitution assessment is a wellness/category classification, not a diagnosis of any specific disease. The report's own disclaimer makes this explicit. Patients with specific symptoms — chest pain, sudden vision changes, persistent fever — go to a hospital, not to a TCM scanner.

The device does not replace the four diagnoses. The pulse-taking, the conversation about sleep and digestion and emotional patterns, the tongue inspection — those still happen at a real consultation with a real practitioner. The device reads the face and the eyes. That is one of four.

The device is not a substitute for a licensed TCM doctor's diagnosis. The report includes a recommendation list — common dietary patterns, common exercise patterns, sometimes herbal teas — but those recommendations follow the constitution type, not a specific person's condition. A user with hypothyroidism on medication is not going to find it on the report, and the report is not going to advise them about it.

China's TCM AI export in context

Guanwei's device is one of three commercially active AI-TCM-diagnostic systems that have come out of China in the past two years. The other two are the Tianjin Da Ren Tang modern TCM project, which uses AI mainly in the manufacturing and quality-assurance stages of herbal products rather than in clinical diagnostics, and a smaller standalone project by Hunan-based Sinocare that targets pharmacy-chain placement for self-service use. The Guanwei unit is the only one with a multi-language rollout specifically engineered for clinical use abroad.

This matters because TCM diagnostic standardisation is one of the bottlenecks China has been trying to break in its 15th Five-Year Plan (2026-2030). The plan's healthcare-technology section calls out TCM AI specifically, asking for "the development of TCM data infrastructure and AI-assisted diagnosis systems that can be deployed at home and abroad." The Beijing conference was, in effect, an early industry checkpoint on that plan line. The fact that three TCM-data papers used Guanwei's data as the validation set suggests the company has a soft monopoly on the high-quality constitution-classification training set — at least for now.

How an international patient can use this today

The device is rolling out to clinics inside Beijing on a partnership basis. As of mid-July 2026, three TCM hospitals and one integrative-medicine clinic in the city centre have it in their consultation rooms. Walk-in appointments are available, but the recommended path is to book ahead through a hospital's international-patient office. The whole scan-and-report step takes about ten minutes; the rest of the consult is a normal TCM consultation.

• Where to go: Beijing-based TCM hospitals with international-patient offices. Within a 30-minute cab of most central Beijing hotels.
• Cost: The scan itself is included in the consultation fee at most locations (RMB 300-800 / US$42-110 for a physician visit). Some clinics charge a separate RMB 200 / US$28 fee for the digital report in PDF/email form.
• Language: English, Korean, Russian, and Japanese reports are available on request. The practitioner consultation is in Chinese with bilingual support at the major hospitals' international offices.
• What to bring: A list of current medications and any recent lab work. The TCM practitioner will want to know about them. The constitution report itself does not ask.

What to watch in the next 12-18 months

Three signals worth tracking.

1. NMPA Class III medical-device clearance. The current device is registered as a Class II wellness device. If Guanwei clears a Class III medical-device review (which would put it in the same bucket as blood-pressure monitors for the home market), that unlocks hospital reimbursement and a much bigger clinical footprint. The company has hinted at a Class III filing by end of 2027.
2. English-language clinical validation. The constitution-classification model has been validated against Chinese-cohort data from the China Academy of Chinese Medical Sciences. A non-Chinese validation cohort — the kind that would matter for an EU CE-mark and a US FDA 510(k) clearance — does not yet exist. Whether one is being assembled is the next data point to watch for.
3. Linkage to TCM hospitals' HIS systems. The killer app for the device is not the standalone scan — it is the scan that flows into a hospital's electronic medical record, where the TCM practitioner already works. Pilot integrations with two hospital HIS systems are reportedly underway. If those land and the report integrates into the same screen the TCM doctor sees for pulse and tongue data, that is the inflection point.

A grounded take

AI TCM diagnostic devices have been promised for at least a decade and have generally under-delivered. What is different about the Beijing v3 is that the engineering has caught up — running inference on-device, multi-language real-time output, the visual quality of the report — and the regulatory framing is realistic (Class II wellness, not pretending to be a clinical diagnostic). For a traveller who is in Beijing and curious about TCM, the device is a reasonable front-end step: it gets you the constitution-type vocabulary you would otherwise have to negotiate in two hours of Chinese conversation, and it gets you to the right herbal-style recommendation in a single visit.

For a hospital or clinic network considering deployment, the questions worth asking first are: how is the constitution model validated against your local patient cohort? what is the false-positive rate for a category the user is on the boundary between? and what is the unit cost at your volume, with how many years of software updates included? Those three answers will tell you whether the device solves a real problem in your clinic or is a marketing exhibit that adds a corner to the waiting room.

Related reading