Restoring a Lost Reading Tradition With AI: Building Jumyeong-myeongri
aimachinelearningllmwebdev
TL;DR
Jumyeong-myeongri is a Korean four-pillars reading service built by Ginigen AI. You enter a birth date and time, it computes an exact four-pillars chart, then reads your character, wealth, work, relationships, health, life cycles, this year, and compatibility. Every reading follows one flow: a sharp, accurate observation, then the chart-based reason, then empathy, then courage and a few concrete next steps. It is free to try.
Link: https://ginigen-ai-saju.hf.space/
Why we built it
Deep reading in this tradition lives with a small number of experts, and the old source material is scattered and hard to reach. We wanted to close that gap with AI: cross-check the classical frameworks and express them consistently in modern language. The core question was simple. Can we automate the calculation and interpretation a human expert goes through, without getting it wrong?
The calendar engine comes first
Everything rests on an exact chart. If the chart is wrong, every sentence above it is false. So the engine computes the pillars from solar terms, the sexagenary cycle, hidden stems, the twelve spirits, noble and inauspicious markers, and rare noble patterns, plus the forward or reverse life-cycle direction. We matched accuracy against a professional reference by hand. The same birth data always yields the same chart, so a chart computed once can be reused.
Writing that people actually read
Flat, textbook prose gets ignored. We designed each section as four beats: a precise observation that makes the reader nod, the chart-based reason, genuine empathy, and then a turn toward strengths and practical steps. No fatalism, no fear selling. The reader also picks a depth: plain language, terms with reasons, or close to the source.
Architecture
The backend is a small API with a single static page. Calculation runs in Python, interpretation goes through an abstracted model backend. Readings are grouped into tabs, and only the tab a user opens is generated at that moment. This lazy generation matters: generating everything up front is slow and costly, while generating only what is viewed lowers the average cost, and identical charts are generated once and shared.
A reply bot on X
To reach more people, we added an auto-reply bot on X. When a user leaves a birth date, time, and sex in a reply, the bot posts a chart summary and a short reading. The hard part is parsing free-form text so real replies are not missed: different ways to write sex, a time with only an hour, or an unknown birth time.
Lessons
- Fix the foundation first. One calculation error turns every sentence into a lie.
- Voice is the product. The same data, read in four beats, feels completely different.
- Lazy generation decides cost. Generate what people actually view.
- Input parsing decides reach. If you cannot read free-form text, a good engine still loses users.
FAQ
Is it accurate? The chart is matched against a professional reference. The interpretation is a modern reading of classical tendencies, for insight and fun.
No birth time? You can select unknown, and time-related parts are read flexibly without hard claims.
Is it free? Yes, you can try it for free.
What languages? Korean for now.
Try it
https://ginigen-ai-saju.hf.space/