FANzR-arch/Numerologist_skills

🔮 An engineering framework to stop LLM hallucinations in Chinese astrology. / 给“赛博半仙”戴上紧箍咒：减少幻觉、固定排盘步骤的奇门遁甲与紫微斗数 AI skills。

What it solves

This project addresses the issue of LLM hallucinations when dealing with traditional Chinese divination systems (such as Qimen Dunjia, Ziwei Doushu, and Bazi). Instead of trying to make a model "more mystical," it focuses on reducing errors in chart casting, school-specific interpretations, and the logical sequence of analysis by modularizing the process into auditable and reusable engineering modules.

How it works

The system decomposes the divination process into three main components:

  1. Prompt Constraints: Controls the order of questioning and the boundaries of the output to ensure the model asks for missing information before generating a result.
  2. References: Fixed sets of rules, glossaries, and interpretation sequences to standardize terminology and logic.
  3. External Scripts: Offloads rigid calculations, such as calendar conversions and chart casting, to Python scripts (e.g., the Qimen Dunjia CLI) to ensure mathematical accuracy.

Who it’s for

Developers and AI engineers interested in the engineering of traditional divination systems who want to create transparent, reliable, and structured AI-driven analysis tools.

Highlights

  • Modular Skill Sets: Includes dedicated modules for Qimen Dunjia, Ziwei Doushu, and Bazi.
  • Hallucination Reduction: Prioritizes external scripts for rigid calculations over LLM generation.
  • Structured Workflow: Enforces a "question first, output second" rule to prevent the model from guessing parameters.
  • Auditability: Uses references and clear declarations of school-specific rules to make the analysis process transparent.

Related

  • Project
    learnwithu/mingli-master

    An AI-powered Zi Wei Dou Shu astrology tool that uses a Python rule engine for precise chart casting and an LLM for human-like interpretation and visualization.

  • Project
    xLLM-AI/xllm

    xLLM is an open‑source C++ inference framework that delivers high‑throughput, low‑latency LLM serving on Chinese AI accelerators (Ascend, Cambricon, Moore Threads, etc.). It separates a brpc‑based service layer from a hardware‑optimized engine, includes hybrid KV‑cache management via Mooncake, and provides ready‑to‑run deployment scripts for models such as GLM‑5.3‑Flash and MiniMax‑M3. The project is Apache‑2.0 licensed and backed by JD.com and several Chinese university labs.

  • Project
    Brhiza/mingyu

    An open-source toolkit for generating precise astrological and divination charts and converting them into structured prompts for AI interpretation.

  • Project
    7toCR/paper2patent

    A collection of professional LLM prompts and templates designed to convert academic research papers into legally compliant Chinese patent applications.

  • Project
    linghungegeg/Linghun

    A local-first, evidence-driven AI programming terminal that integrates LLMs into real projects with system-level constraints to reduce hallucinations and improve engineering reliability.