Using AI to conduct Tibetan medicine constitution identification, Zangyi Xiaozhuoma aims to get the "assessment-solution-product" closed loop fully operational on the HarmonyOS portal.
For constitution identification, Tibetan medicine has its own theoretical basis: the prosperity, decline and imbalance of the three factors Lung, Chipa and Pekan correspond to a complete system of methods ranging from diet and daily routine to external treatment and conditioning. However, in daily health services, this set of judgment logic has long depended on the inspection and pulse diagnosis experience of veteran experts, which is hardly accessible to ordinary users. Most health products on the market follow a different routine — users fill in a questionnaire and get a vague generic suggestion, which is completely disconnected from the syndrome differentiation logic of Tibetan medicine.
The practice of Zangyi Xiaozhuoma is to turn the three-factor theory of Tibetan medicine into a computable system for constitution identification and scheme generation, and then reach users through the system-level entry of HarmonyOS. The project has been launched for internal beta test, and is currently running small-scale trial operation at the Yan'an Health and Wellness Base. It plans to start angel round financing after verifying the model through trial operation.
"Tibetan medicine believes that the imbalance of the three factors, Lung, Chipa and Pekan correspond to different states of Qi, energy and body fluid respectively, and this classification is the foundation for Tibetan medicine practitioners to judge constitutions and formulate conditioning schemes," said Ma Wen, the person in charge of the project. The first step the team took was to break down this classification into quantifiable evaluation dimensions: physical features, physiological state, living habits, personality tendency, plus the data of heart rate, sleep and activity volume from wearable devices. The model will output the judgment of constitution type, and generate corresponding schemes for diet, daily routine, medicinal bath and physical therapy.
The difficulty does not lie in the model generating text, but in the boundary control. Health-focused AI is most frequently asked questions like "can it make diagnosis" and "does it have curative effect". Zangyi Xiaozhuoma's solution is to strictly limit its business scope within health suggestions and conditioning services: it does not make diagnosis conclusions, does not promote curative effects, does not describe health supplements as medicines, and at the same time adds explicit and implicit marks to AI-generated content in accordance with the latest requirements, and implements separate authorization and full-link encryption for health file data.
The direct benefit brought by this boundary design is that the closed-loop of re-test can be established. After the user gets the scheme, the data of the next evaluation will be fed back to the model to adjust the scheme suggestions for the next round. The more data is used, the more accurate the model will be, which is the key to the standardization of highly individual experience-dependent services such as Tibetan medicine.
What really makes this project stand out is the positioning it aims for. The HMAF 2.0 framework of HarmonyOS supports mutual calls between Agents through the A2A protocol. Developers can register their services as system-recognizable Skills, which can be directly invoked by system-level agents such as Celia after understanding the user's intention. This means that Zangyi Xiaozhuoma does not need to purchase traffic for promotion. As a Skill for constitution identification and health conditioning, it can be embedded in the interaction chain when users talk to their mobile phones.
Compared with an independent App or mini-program, the difference of this system-level entry is that users do not need to perform extra download operations, and the service can be directly integrated into users' existing usage habits. HMAF 2.0 has opened more than 20 AI capabilities and GUI control capabilities, which is a new traffic distribution window for service providers in vertical scenarios.
In terms of business model, Zangyi Xiaozhuoma builds the whole process from evaluation to scheme into a commodity export: after constitution identification, users can purchase corresponding food-medicine homology commodity packages, or access course-based physiotherapy services such as medicinal bath and Hor moxibustion. In addition, there are membership subscription and SaaS services for wellness bases. The internal calculation of the project shows that the payback period of a single wellness base is about 3 to 6 months after the process from scheme to commodity is fully implemented.
The cold start of the project is also based on physical entities. The Yan'an Health and Wellness Base provides landing scenarios for the first batch of user pools, and allows the model to obtain training data feedback in the process of serving real users.
There are many similar practices in the industry. The Tibet Autonomous Region has promoted the digital construction of the Tibetan medicine calendar and arithmetic resource center and the Tibetan medicine constitution identification system. Multiple institutions and hospitals in Shenzhen and Tibet are also developing digital tools for Tibetan medicine constitution identification. In the 2026 list of benchmark demonstration projects of Lhasa, there are large Tibetan medicine models and intelligent auxiliary diagnosis and treatment projects for plateau health services. On the HarmonyOS side, relevant promotion is also underway: Longgang District has already launched a full-scenario open-source HarmonyOS smart nursing home, which connects smart devices to the same HarmonyOS base.
These practices show that the digitalization of Tibetan medicine and HarmonyOS-based health and wellness services are both directions being promoted by their respective industries, but the practice of connecting the two and simultaneously linking to commodity supply has not yet formed a mature sample.
There are still several hurdles to cross from verification to large-scale implementation. The exclusive authorization of experts, software copyrights and data asset registration determine the thickness of the moat of this set of knowledge assets; the re-test rate and scheme implementation rate in the trial operation stage determine whether the data flywheel can really rotate; whether the step of converting schemes into commodities can be realized directly determines whether the business model is feasible.
The project is currently in the internal beta test and trial operation stage. The next step is to complete the first round of trial operation at the Yan'an base, verify the model and business link with real user data, launch angel round financing, supplement the product and compliance teams, and then replicate the model to other wellness bases in the northwest region.
Project Information: Zangyi Xiaozhuoma | Positioning: AI Service for Tibetan Medicine Three-Factor Constitution Identification and Health Conditioning | Founder: Ma Wen | Stage: Internal Beta Test and Trial Operation | Financing: Plan to Launch Angel Round | Location: Yan'an, Shaanxi