Skip to content

What we take on

Five lines below, but it isn't a menu to order from. We listen to what you're already doing and work out which line it sits closest to. Most of the time it sits across two.

Ruled this far and stopped.

What we take on

Building past launch

The part after launch is written into the contract.

The first build, running it, and what gets opened the quarter after all go into one contract. Once it is live we read the numbers together: where people drop off, what would make them stay.

We write the next quarter's list first and bring it in, having looked at what is getting a good response in that industry lately. A few quarters in, the questions change. What used to be "can this be done" becomes "what are we opening this quarter." How far it goes still gets written down again each quarter.

Where it fits
Places where nobody has decided who watches this after it ships. Places where re-tendering every next step is more trouble than it's worth.
If you want to start small
One thing you're already running becomes a one-quarter contract. At the end of it we write the next list together, and you say there whether it continues.

Evidence: Commerce operations tool, Ad channel performance consolidation

AI agents, built and run

The stops go on before it ever runs.

Tools get attached and the loop runs. The brakes go on before that. What the brakes actually are: how much it can spend in one go, how many times it may retry the same thing, how many seconds one call gets before it is cut.

The editing agent we are building has 31 tools and 5 brakes. It ships switched off, and a person is the one who switches it on.

Where it fits
Places where a person makes the same judgement several times a day. And places that can say what goes wrong when that judgement is wrong.
If you want to start small
We pick one repeated judgement and wire that up. It stays with a person confirming the last step, and after a few weeks of watching it run, that step comes out.

Evidence: Story editing studio

Commerce and marketing ops

The work you were doing across a dozen tabs comes down to one screen.

Orders, waybills, and customer questions that used to mean logging into each marketplace separately get handled in one place. Same for ads: spend and results scattered across channels come into one table. These are things we built for our own use and keep running. On top of that goes the way the company already works.

The advertising operations automation for BK파트너스 is running on this line right now.

Where it fits
Sellers past their third channel. Places where the day goes into moving orders from one screen to another.
If you want to start small
One channel goes in first. Run it for a month, see how much hand-work it took off, then the rest follow.

Evidence: Commerce operations tool, Ad channel performance consolidation

Rebuilding what already runs

Swapped out without going down.

What's running now gets read first: where the data sits and how it got there, which screens are actually opened and which ones aren't. The order comes after that.

Plans to replace everything at once mostly stall halfway. The screen people open most often gets rebuilt first, and old and new run side by side for a while. Screens nobody opens don't get carried over. That's the most common place a rebuild quote swells.

Where it fits
Systems whose original builders left long ago. Places that have left it alone because nobody knows what gives when you touch it.
If you want to start small
One screen gets rebuilt first. Watch that one run alongside the old system, then the scope gets set.

Evidence: Multi-brand web, Redevelopment information map

Foundations and nonprofits

It has to keep running after the person in charge changes.

Intake for counselling, and the screens that carry those records into the reporting a programme has to file. Here the budget is tied to a fiscal year and the numbers land in the closing report as they are. So what gets settled before the build is the remaining budget and the shape of the report. Leaving it readable for whoever comes next is for the same reason.

The counselling platform and website for 삼선배움과나눔재단 were taken on directly in 2024. The Save the Children web service ran for a year, through 디나인컴즈.

Where it fits
Programmes with a budget attached that has to be closed out inside it. Places where next year's person in charge may not be this year's.
If you want to start small
Setting the scope and the figure for the programme proposal is the first contract. Starting the build after that clears is not too late.

If you can't tell which line it is, just write down what you're trying to do. Working out which line it is, is our job.

Questions we get

What does handover look like when the build is finished
Every deliverable hands over as a documented repository, a reproducible local setup, and a migration history. All schema changes exist as migration files, so local and production reproduce in the same order. The runtime is pinned in files such as a Docker Compose definition, which lets the receiving side bring up the same environment without a verbal walkthrough.
Can you take over a system that is already in production
A system already in production starts with an architecture review. The existing codebase and infrastructure are read, and bottlenecks, data-model constraints, operational risks, and a repair priority are written down with reproduction steps. Once that document exists, the scope that can safely be touched in one pass is agreed, and implementation proceeds inside that scope.
Can we start from an idea without a written spec
A written spec is not required to start. The first conversation works out what can realistically be built in one pass, and that write-up takes the place of a spec. Telling us what is already decided and what is still open is enough to begin narrowing the scope.
Do you train models, or use provider APIs
Provider APIs cover most of the work, with local models used only where they are needed. For image generation and video processing the fall-back route is decided in advance, and the bottom of it is fixed to a path that completes without a GPU. For text generation, providers sit behind a protocol and that is as far as it goes today.
Can the work be conducted in English
Work can be conducted in English. Every page of this site exists in both Korean and English, and project documents, code comments, and commit messages are written in English as well. The language for replies and working documents is agreed at the time of the inquiry.

We start by settling how far you want to hand over.

Email
info@divine-company.co.kr
Phone
070-8955-2815
Address
Seoul Seongbuk-gu 713 Bugaksan-ro, Unit 3-307