Notes on shipping a four-language fortune-telling product without a co-founder, an investor deck, or a permission slip.
There is a particular kind of loneliness in building software alone. You wake up, look at the dashboard, see the number, and decide whether the number means today is real or not.
For me the number is users of a Korean astrology SaaS that I built and run by myself, in four languages, with payments wired through PayPal and Korean PG, with a recommendation engine that turns a person's birth time into a year-long narrative — and with no team to share the wins or the silences with.
This is the story of how I got here, what I wish I had known eighteen months ago, and why I think building an AI-native product around a thousand-year-old tradition is not a contradiction but an opportunity that almost no one is taking seriously enough.
The wedge: a tradition that already has demand
Astrology as a global category is not a fringe market. The most recent third-party research I trust pegs the astrology app sector at roughly $5.7B in 2026, on a path to $11.7B by 2030 at a compound rate near 20% per year. That is not a niche. That is a category bigger than most "AI for X" pitches you see on a typical demo day.
Inside that category, the West has built a confident generation of horoscope products powered by sun signs, NASA ephemeris data, and a moody UI voice. What it has not built — and what has been almost entirely overlooked — is a serious productization of Eastern astrological systems: the Korean and Chinese tradition of saju (the four pillars of destiny), the Vedic chart, the ten heavenly stems and twelve earthly branches, the elemental compatibilities that govern marriage, business partnerships, and timing in cultures that still take these calculations seriously when buying a house.
I am Korean. I grew up watching my parents pay a human master a hundred thousand won to read someone's chart by hand. The math is deterministic. The interpretation is not. That gap — deterministic input, narrative output — is exactly the shape of work that a modern language model is good at.
That was the wedge.
What I built
The product is a Korean astrology and life-pattern SaaS. The technical spine looks something like this:
- A birth-time engine that converts solar time, lunar calendar, time zone, and true-solar-time correction into the canonical four-pillar representation. This is the part no Western astrology app ships, and it is non-trivial because lunar calendar tables have edge cases that bite.
- A generation layer that feeds the pillar representation into a language model with a tightly bounded prompt scaffold. The model never gets to invent the chart. It only narrates the chart it is given.
- A four-language front end. Korean and English are the obvious two. Japanese and Chinese were added next because the underlying tradition is shared and the cultural translation cost is much lower than re-inventing the whole interpretation tree for, say, German.
- A payments layer that supports both global subscriptions and a Korean PG so that someone in Seoul can pay with their everyday wallet and someone in Los Angeles can use a global card.
- A recommendation engine that compares a chart against curated themes (career year, relationship year, health year) and produces a personalized PDF in the user's language.
It runs. It accepts money. The hardest part was not the model. It was the calendar.
What it cost in time
Eighteen months, mostly evenings and weekends at the start, full attention for the last six. Around two thousand commits across the monorepo. Two complete rewrites of the interpretation layer when I realized the first version was hallucinating in a way I could not bound. One pivot of the brand voice. Three pricing experiments. Many quiet weekends.
What it did not cost was a team. I have never had a co-founder. I have never had an employee. The closest thing I have to colleagues are the AI executives I run internally — but that is a separate essay and I will not collapse the two threads here.
What I wish I had known
A few things, in the order they would have helped me most:
One. The hardest single decision in this kind of product is not "which LLM" — it is what the model is allowed to invent versus what it must repeat verbatim. For a system rooted in tradition, "must repeat verbatim" needs to win more often than your intuition will tell you. Users do not want a model improvising on the meaning of the year of the Red Horse. They want the canonical interpretation rendered in their language, in their tone, for their chart. Inventiveness is a bug here, not a feature.
Two. Localization is not translation. The Japanese reader expects the chart to be presented through the lens of jūnishi and jikkan, with vertical-friendly text rhythm. The English reader expects a narrative arc that builds toward an actionable summary. The Chinese reader expects denser information density per screen. None of this is captured by a translation API. All of it has to be designed.
Three. A solo founder needs two flywheels, not one. The first is the product. The second is the infrastructure for shipping the product without supervision: scheduled tasks, monitoring, alerting, retry queues, and a daily morning brief that tells you whether the night-shift code did its job. Without the second flywheel, the first one stops every time you take a weekend off, and a year later you have shipped six months of work.
Four. The honest pricing answer for an Eastern astrology SaaS is not "match Western pricing." Western horoscope apps anchor at three to ten dollars a month. Eastern astrology has a history of one-time, high-value readings — the in-person master charge is typically five to fifteen times higher than a monthly horoscope subscription. There is room here for an annual plan, a family plan, and a premium consultation tier that Western competitors have not even considered.
Five. Do not chase paid acquisition before the product can hold the user past the first chart. The category is generous to organic. Pinterest, TikTok, long-form blog content, Bluesky, Hashnode, and Mastodon will quietly feed you traffic that converts better than any cold ad — if the first reading is good enough to talk about at dinner.
On being alone
People ask me whether building a SaaS alone is harder than building one on a team. The honest answer is: it is differently hard.
A team gives you debate, redundancy, and a calendar full of other people's energy. A solo build gives you decision speed, full context on every line, and zero meetings. The trade is real. I have shipped features in twenty minutes that would have been a two-week sprint at a startup with a stand-up culture. I have also gone three days without speaking to a human about the work, and that part is corrosive if you do not engineer around it.
What works for me is to treat the company as if it were a real company, even though it is not. Department names. Daily reports. Weekly reviews. Internal letters between fictional teams. A small fleet of AI executives that hold the work I would have otherwise dropped on the floor.
That second flywheel — the operating system for a one-person company — is what makes the first flywheel survive.
What is next
I am not yet at the scale where this product is generationally important. I am at the scale where it is real, where strangers pay for it, where it speaks four languages, and where the question is whether I can ten-x it without breaking the trust of the people who already paid.
The category is large. The Eastern astrology corner of it is wide open. The technology is finally good enough to render a tradition without flattening it. And the cost of one person building this in 2026 is genuinely the lowest it has ever been.
If you are sitting on a domain that has demand, a tradition, and a math that a model can carry — and you have been waiting for a co-founder before starting — consider that the co-founder you have been waiting for is the second flywheel.
Build it. Then build the first.
This is the first in a three-part series on building an AI-native Korean astrology SaaS as a one-person company. Next: how four AI executives run the operations of a company with zero human employees.