Your LLM just enrolled in Chinese medical school

What it does

nihaixia is a knowledge-packaging project masquerading as a repository. It distills 2,452 pages of lecture notes from the late TCM master Ni Haixia into markdown modules and IF-THEN diagnostic rules consumed by AI agents via the OpenClaw skill format. Once loaded, a compatible LLM can attempt Six Meridians diagnosis, classical formula selection, and pulse-and-tongue pattern matching drawn from 3.5 million characters of source material.

The interesting bit

Instead of treating classical medical texts as a simple retrieval corpus, the project encodes them as a reasoning framework. Version 2.1.0 introduced explicit “diagnostic formulas”—decision trees and a seven-step thinking pattern that force the model to follow Ni’s clinical logic rather than freestyle hallucination. A dedicated voice module even nudges the LLM to reply in the master’s blunt, spoken cadence.

Key highlights

• Nine markdown modules covering the Shang Han Lun, Huang Di Nei Jing, Shen Nong Ben Cao Jing, 849 clinical cases, and acupuncture theory.
• Eight Six-Meridians diagnostic formulas with flowcharts, pulse/tongue quick-reference, and contraindication rules meant to constrain LLM reasoning.
• Includes Ni’s non-medical teachings: Purple Star astrology, I Ching, and feng shui from the “Tian Ji” lecture series.
• Packaged for OpenClaw and Claude Code; a separate offline Android app reuses the same knowledge base.
• Self-reported validation cites a 9.4/10 “Darwin skill score” and 100% formula-dose accuracy against source files.

Caveats

• This is a content-curation and prompt-engineering artifact, not a medical device or clinically verified diagnostic system; it requires an external AI agent platform to be useful.
• Astrology, fortune-telling, and feng shui modules sit alongside medical content, which may surprise users looking for a purely clinical resource.
• The “skill” format is essentially a large, structured prompt with embedded reference material—there is no standalone inference engine or traditional application code here.

Verdict

Worth exploring if you are building domain-specific AI agents or studying how classical expertise can be compressed into constrained LLM reasoning. Look elsewhere if you need peer-reviewed medical software, a standalone app, or executable diagnostics—this is a syllabus with guardrails, not a stethoscope.