Speak normally
You can describe the problem in ordinary language without technical vocabulary or a formal specification.
JP
JP
I did not start out knowing programming or web development. While building models, I described things I wished existed to ChatGPT, made rough prototypes, tested them, and brought the failures back for revision. The tool-making side of 299 MODEL WORKS grew out of that dialogue.
This page is not a claim that ChatGPT automatically completed everything. It is a record of repeated dialogue, real-device testing, and revision toward something usable.
00 / WHAT IS GPT
On this page, “GPT” means ChatGPT, which I use as a thinking and development partner for 299 MODEL WORKS.
ChatGPT is an AI assistant you can work with through conversation. It can help answer questions, organize ideas, draft text, plan work, write code, and inspect images or files while the discussion continues.
I see it less as a machine that produces a finished answer and more as a partner that organizes goals and constraints from what I explain, then proposes the next thing to test.
You can describe the problem in ordinary language without technical vocabulary or a formal specification.
A vague idea can be separated into goals, conditions, and steps.
It can produce a first draft of text, code, or interface ideas that can be run and checked instead of only discussed.
GPT is not always correct. For work such as scale models and 3D data, where the outcome can be checked physically, you still need to test it yourself and revise what does not work.
02 / DIVISION OF WORK
03 / WHAT GREW
It began as a tool for finding model paints close to a selected color. Through actual use and revision, it has gradually expanded into a broader way to find the paint you need for a model.
Small add-ons for inspecting, selecting, splitting, and repairing 3D data used in model making. One was added each time a repeated problem appeared; there are now 13.
A calculation helper for sprocket dimensions, center distance, and chain link count. It grew from a rough idea into a tool that also supports CAD work and SVG output.
This site itself — collecting finished models, tools, and build records — was also built by comparing prototypes and correcting one awkward detail at a time.
04 / THE LOOP
Notice a repetitive, tedious, or failure-prone step during model making.
Without writing a formal specification, explain in ordinary language what is frustrating and what would make the work easier.
Do not aim for a finished product immediately. Build the smallest working version and use it on real data.
Show the exact data that fails, the interface that feels wrong, or the process that is too slow, then look for the cause.
Once it becomes useful, preserve it as a tool, procedure, or article for the next project.
05 / REALITY CHECK
GPT does not erasewhat I do not know.It helps me finda way to move forwardeven while I am still learning.
It can produce convincing mistakes: nonexistent functions, incorrect assumptions, or interfaces that are impractical in real use.
With scale models and 3D data, the final result exists in the real world. You cannot judge it without testing on actual data and physical output.
For that reason, 299 MODEL WORKS treats GPT suggestions as prototypes rather than finished answers. I use them, go back when something is wrong, and ask again when something is unclear.
Nothing becomes finished instantly. But by checking one unknown at a time, testing, and revising, I can now carry an idea far enough to use it in my own work. Paint Finder, the Blender add-ons, and Chain Builder all grew from this loop.