TCM Health AI Assistant

Outline

The TCM Health AI Assistant leverages advancements in large language models (LLMs) to modernize traditional medicine practices across Asia, with a particular focus on Traditional Chinese Medicine (TCM) and Kampo. By integrating AI with traditional medical knowledge, the system offers personalized health recommendations and treatments. It aims to enhance the accessibility and effectiveness of traditional medicine for both clinical applications and consumer health management.

This project seeks to revolutionize how traditional medicine is understood and applied in the digital age by blending ancient wisdom with cutting-edge AI technology.

Inspiration

The TCM Health AI Assistant was inspired by the rapid advancements in AI, particularly LLMs, and the rich, holistic traditions of TCM and Kampo. We saw an opportunity to bridge the gap between ancient medical practices and modern healthcare by offering a system that makes traditional treatments more accessible and effective. Our aim was to create a tool that could benefit both healthcare professionals and everyday users seeking holistic health solutions.

By integrating ancient knowledge with modern AI capabilities, we hope to revolutionize the way traditional medicine is applied and perceived in the modern world.

What it does

The TCM Health AI Assistant provides personalized health assessments based on TCM principles. It evaluates user symptoms (focusing on the four diagnostic methods: observation, smell, questioning, and pulse-taking), body constitution, and lifestyle factors to recommend Kampo formulas, dietary suggestions, and holistic treatments. Additionally, it draws insights from historical TCM texts, contextualizing ancient knowledge for modern use. The system also supports clinical diagnoses and remote health consultations, offering a versatile tool for both professionals and consumers.

How we built it

We built the TCM Health AI Assistant using a combination of Gemini 1.5 Pro and Dify Rag systems, which were trained on traditional Chinese medical texts, modern clinical data, and AI-driven natural language processing models. Our multi-disciplinary team collaborated to translate the holistic, non-linear TCM diagnostic approach into a structured format that AI could understand. Advanced algorithms were used to extract valuable insights from historical Chinese texts and adapt them for contemporary medical practices.

Challenges we ran into

One of the biggest challenges was converting the complex and often non-linear diagnostic methods of traditional medicine into a format that AI could process. TCM diagnoses are highly personalized, with treatments varying based on symptoms, patient constitution, lifestyle, and emotional factors. Maintaining the depth and nuance of TCM while ensuring AI accuracy was a significant hurdle. Additionally, integrating older, often ambiguous texts into the model required deep linguistic processing and contextual understanding.

On the technical side, we faced challenges in balancing data and model fine-tuning with Rag, as well as determining the optimal granularity for chunking in Rag indexing.

Accomplishments that we're proud of

We successfully developed a system that bridges the gap between ancient traditional medical practices and modern AI technology. Our model’s ability to suggest personalized treatments, rooted in TCM’s complex diagnostic methods, marks a major breakthrough in healthcare AI. We are particularly proud of the educational content the platform generates, as it demystifies TCM for a broader audience and brings ancient healing practices into contemporary health discussions.

What we learned

Firstly, this project showed us the immense potential AI has for transforming not only modern medicine but also traditional medical practices. We learned the importance of cross-disciplinary collaboration between AI experts and traditional medical practitioners. Most crucially, we realized how essential it is to preserve the integrity of traditional methods while making them accessible through modern technology.

Secondly, after integrating Gemini 1.5 Pro as the LLM provider, we saw a significant improvement in the accuracy and professionalism of the TCM Health AI Assistant, especially in pattern identification and treatment determination (弁証論治). This shift from using local LLama 3.1 highlighted the importance of choosing the right LLM provider, which was my most valuable discovery during this hackathon.

What's next for TCM Health AI Assistant

Moving forward, we plan to expand the diagnostic capabilities of the TCM Health AI Assistant and introduce more advanced personalized treatment plans. I am continuing to collaborate with Chinese medicine specialists to validate the treatments using extensive TCM case data. Additionally, I plan to incorporate more TCM texts and theories from renowned TCM experts into the Rag system to further enhance its accuracy and usability.

Built With

• dify
• gemini
• nomic-embed-text
• ollama
• rag