Page 128 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Key Aspects of 1. Using embedding technology, we developed a system for parsing, stor-
Solution ing, and retrieving multimodal data. By converting TCM prescriptions,
literature, and related materials into vector representations, we built a
multimodal (e.g., text, images, PDFs) vector knowledge base, effectively
addressing challenges in data utilization.
2. By incorporating KooSearch-enhanced retrieval, we have developed core
capabilities such as query rewriting, post-retrieval risk validation, citation
localization, and multimodal (text-image) match positioning—effectively
reducing hallucinations in large language model outputs.
3.Integrating large models and small models, we developed a multimodal
RAG platform for fast, scalable AI assistant deployment across diverse use
cases.

Technology Digital foundation, knowledge graphs, Vector database, Large Language
Keywords Model

Data Availabil- TCM prescriptions, TCM literature, Classical TCM Texts, Chinese Pharmaco-
ity poeia, Phytopharmaceutical Study Datasets, Labeled TCM Practice Data

Metadata text, image
(Type of Data)

Model Training Huawei Pangu's Large Language Model solution provides an AI computing
and Fine-Tun- infrastructure, AI development platforms, and tools to construct a vector
ing knowledge database. Yunnan Baiyao collects high-quality datasets, trains and
fine-tunes them on the Pangu Large Model to develop the Lei Gong Large
Model. By integrating prompt engineering, it provides comprehensive appli-
cation services: supporting farmer cultivation, accelerating product R&D, and
delivering health management solutions.
Testbeds or Leigong Large Model for Traditional Chinese Medicine (TCM)[5]
Pilot Deploy-
ments

Code reposi- Not available
tories

92