Page 131 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

REQ-02: Yunnan Baiyao leverages its compiled high-quality datasets to train and fine-tune the
Pangu Large Model, developing the Lei Gong Large Model specialized for the TCM industry.

REQ-03: By integrating the TCM vector database with prompt engineering, the system mitigates
hallucination during incremental training and inference processes of large models. 4.1-Healthcare

REQ-04: Expert review and annotation: During the knowledge database construction phase,
only authoritative source content is selected, such as official authoritative texts like the "Chinese
Pharmacopoeia" and "Chinese Herbal Medicine." A team of TCM experts annotates key
information such as the efficacy of medicinal materials and contraindications, and establishes
a "confidence level grading" system.

REQ-05: Quality control of model training and output: A dual review mechanism is used
to evaluate the output results. First, machine filtering of keywords is employed, such as
automatically blocking high-risk responses with absolute statements like "cure" and "anti-
cancer". Second, manual spot checks are conducted, with 5% of the generated results randomly
selected daily for evaluation by professional Chinese medicine practitioners.

REQ-06: Adhering to China's "Good Agricultural Practices for Chinese Medicinal Materials
Certification Management Measures" and "Interim Measures for the Management of Generative
Artificial Intelligence Services." The Leigong large model complies with requirements for
data security, content accuracy, and qualification filing. In data security, all user health data
is anonymized; for content accuracy, content is reviewed by a team of traditional Chinese
medicine experts, RAG technology is used to reduce LLM hallucinations, output results are
marked with content sources, and labeled as "Content is for reference only and does not
replace professional diagnosis and treatment." For qualification filing, it is stated that the current
version is a tool to assist doctors, not a replacement for diagnosis, and does not require medical
qualification filing.

4 Sequence Diagram

Based on data from traditional Chinese medicine, prescription libraries, and traditional Chinese
medicine literature, transform it into vector data to construct traditional Chinese medicine
vector data.

Initiate knowledge and business inquiries on the platform, first review the content and plan the
search. Then use the Pangu search function to achieve natural language vector retrieval. For

95