2026-03-02|7 min read|--AI--Build in Public--Debugging--Architecture--LaKinh

The Day My Fortune-Telling App Started Lying to Users

## The Bug That Shouldn't Exist

It's 10 PM. I'm staring at two fortune-telling charts that should be identical.

They're not.

One says "Tử Vi star is in Dần palace." The other says "Tử Vi star is in Thân palace."

That's 6 palaces apart. In a 12-palace system.

For context: In Vietnamese astrology (Tử Vi Đẩu Số), getting the Tử Vi star wrong is like building a house and placing the foundation in the wrong city. Everything else collapses.

And my app — LaKinh — has been confidently placing it in the wrong palace since day one.

## How Did We Get Here?

Let me rewind.

Three months ago, I decided to build LaKinh: a Tử Vi fortune-telling app. Not because I'm superstitious (I'm a designer-turned-developer, remember?), but because I saw a gap.

The gap: Every Tử Vi app uses ONE school of calculation. But there are THREE major schools:

  • >Trung Châu (中州派) - Hong Kong style, most popular
  • >Nam Phái (南派三合) - Traditional Taiwan, 108 stars
  • >Bắc Phái (北派欽天) - Ancient imperial, flying stars

They calculate differently. Sometimes drastically.

Think of it like programming languages. JavaScript vs Python vs Rust. Same goal (making software), different philosophies.

My vision: Build the first app that supports ALL THREE schools. Let users compare. Let them decide.

Bold vision. Terrible execution.

Because first, I had to get ONE school working correctly.

## The Test That Changed Everything

Today, I decided to test LaKinh against Kabala.vn (a popular Vietnamese Tử Vi site).

Test case: Trương Bá Chi (張柏芝)

  • >Born: May 24, 1980, 7:33 AM
  • >Gender: Female

I plugged the data into both apps.

LaKinh said:

Cung Mệnh: Sửu
Tử Vi star: Thân palace
Thiên Cơ star: Mùi palace
Thiên Lương star: Sửu palace (life palace)

Kabala said:

Cung Mệnh: Sửu ✅
Tử Vi star: Thìn palace ❌
Thiên Cơ star: ??? ❌

Wait. They agreed on the Life Palace (Cung Mệnh). But the 14 major stars? Zero matches.

Not "slightly off."

Zero. Out. Of. Fourteen.

## The Panic Phase

As a designer-turned-developer, my first instinct was: "I broke something."

My second instinct: "Maybe I broke EVERYTHING."

I pulled up my conversation with Kai (my AI coding assistant — one of the 5 AI tools I use every day). We'd been building LaKinh together for weeks. Kai handles the complex calculations, I handle the design and user experience.

Me: "Kai, we have a problem. LaKinh doesn't match Kabala. At all."

Kai: "Let me check the calendar conversion first."

Kai verified: The lunar calendar conversion was perfect. The Can-Chi (天干地支) calculation was perfect. The Life Palace (Cung Mệnh) calculation was perfect.

But the 14 major stars? All wrong.

The culprit: The getTuViIndex() function.

The function that places Tử Vi — the MOST IMPORTANT star — in its palace.

I had implemented it based on... honestly, I don't remember. Some forum post? A half-translated PDF? The fog of "ship fast, iterate later"?

It was completely, mathematically, embarrassingly wrong.

## The Fix (With A Plot Twist)

Kai: "Let me compare LaKinh with iztro."

iztro is an open-source Tử Vi calculation library. Well-tested. Community-verified. The gold standard for Trung Châu school calculations.

Result: 0/14 stars matched.

LaKinh wasn't even calculating the same SCHOOL correctly.

The fix: Replace my DIY algorithm with iztro's proven formula.

Time to fix: 30 minutes (Kai wrote it in Claude Code CLI)

Result after fix: 14/14 stars matched iztro. Perfect.

But here's the plot twist:

When we went back to compare with Kabala... Kabala was also wrong.

Or more accurately: Kabala was using a DIFFERENT school. Possibly Nam Phái (southern school) instead of Trung Châu (middle school).

Which meant:

  1. >LaKinh was broken (fixed ✅)
  2. >Kabala wasn't wrong, just using different rules
  3. >My original vision — supporting multiple schools — was MORE important than I thought

Because even the "authoritative" sources don't agree!

## The Refactor (26 Minutes to Glory)

With LaKinh finally calculating correctly, Kai proposed something ambitious:

Kai: "Let's refactor the entire engine. Multi-school architecture. Right now."

Me: "Right now? It's 10 PM."

Kai: "Right now."

The plan:

  1. >Create a SchoolEngine interface
  2. >Extract current logic into TrungChauEngine
  3. >Build a TuViCalculator facade that can switch schools
  4. >Make it extensible for future schools

Time estimate: 2-3 hours.

Actual time: 26 minutes and 57 seconds.

Kai (Claude Code CLI) churned out:

  • >8 new files
  • >Clean interfaces
  • >Backward compatibility
  • >Passing tests

I just watched. Occasionally saying "yes" or "that looks good."

This is what I mean by "designer lost in code land."

I can't write this kind of code at 10 PM. But with Kai? We ship production-ready architecture in under 30 minutes.

## What I Learned Today

### 1. Never Trust Your First Implementation

Especially when dealing with domain knowledge you don't fully understand.

I'm not a Tử Vi master. I'm a designer who thought "how hard can ancient astrology be?"

Answer: Harder than React state management.

### 2. AI as a Safety Net

Kai didn't just write code. Kai:

  • >Caught my bugs
  • >Researched the correct algorithms
  • >Explained why my approach was wrong
  • >Refactored the entire architecture
  • >Made it extensible for the future

This is the AI workflow I've been evangelizing. Not "AI does everything." But "AI makes your ambitious ideas actually achievable."

### 3. Build in Public Means Showing the Bugs

I could've hidden this. Quietly fixed it. Pretended LaKinh was always correct.

But that's not the point.

The point is: Solo makers make mistakes. What matters is how fast you catch them.

Today I caught a critical bug. Fixed it. And refactored into something better than before.

In one evening.

That's the superpower.

### 4. Multiple Sources of Truth = Feature, Not Bug

The fact that different Tử Vi schools calculate differently? That's not a problem to solve.

That's the PRODUCT.

LaKinh will let you:

  • >Choose your school (Trung Châu, Nam Phái, Bắc Phái)
  • >Compare two schools side-by-side
  • >Understand WHY they differ
  • >Decide which you trust

Nobody else is doing this.

Because it's hard. And confusing. And most apps want to be "authoritative."

But I'm a designer. I know users don't want authority. They want understanding.

## What's Next for LaKinh

This Week:

  • >Document Trung Châu school (done ✅)
  • >Implement Nam Phái school (108 stars instead of 50)
  • >Test against Kabala's "Nam Phái" mode

Next Week:

  • >Implement Bắc Phái school (flying stars logic)
  • >Build the comparison UI
  • >Beta testing with real users

By End of Month:

  • >Launch: "The world's first multi-school Tử Vi app"
  • >Blog series: "How each school differs and why"
  • >Open-source the engine (maybe?)

## The Meta Lesson

Here's what today really taught me:

Building with AI isn't about replacing skills. It's about AMPLIFYING ambition.

Three months ago, building a Tử Vi app seemed hard but doable.

Today, building a multi-school Tử Vi app — something that requires deep domain knowledge in THREE different calculation systems — is ALSO doable.

Same timeline. Same solo developer. Just with AI.

That's the unlock.

Not "AI writes code for you."

But "AI makes your impossibly ambitious ideas suddenly possible."

## For the Curious

If you're interested in:

  • >How the Tử Vi calculation actually works
  • >The architecture of the multi-school engine
  • >The differences between the three schools

I'm planning a technical deep-dive series. Subscribe to my newsletter if you want it. (Or just build your own using Claude Code CLI — the tools are there.)

For now, I'm going to sleep.

Tomorrow, I teach an AI to calculate fortunes in THREE different ancient systems.

Because apparently, that's my life now.

Follow the build: Twitter/X @gianghaison | Newsletter: The AI Builder

Try LaKinh (when it's ready): lakinh.vn

P.S. If you're a Tử Vi master and I got something wrong... please tell me GENTLY. I'm just a designer trying my best. 🙏