You don’t write code to build a decision tree with AI in SMEme. You build it in the chatbot you already use, with prompts SMEme gives you on its quick start page. °FAI: How to Build a Decision Tree.
Part 1 covered why tribal knowledge doesn’t stick in a prompt. Here, I’ll show how SMEme captures it.
Three ways to start.
If you’ve written the procedure down, you attach it. The AI turns it into a tree.
If it lives in your head, you have the AI watch you work a real case. It asks why every time you make a call that isn’t obvious. It notes where you got each fact. Then it builds the tree from what it saw.
If you don’t know where to start, it interviews you and suggests up to five decisions worth encoding.
The second one fits our industry. Most good separators never wrote down what they know.
You approve it three times.
The AI stops for your sign-off three times. First the outcomes. Then the questions. Then the full outline. It saves nothing until you approve.
Take an underbase call. In my pipeline, the separation tool knows what each factory needs for underbasing on different colors. You can assign it by color or by luminosity. As a tree, you might set three outcomes: full underbase, partial, or none. You’d ask about garment color, ink color, and fabric.
You add a note to each question saying where the answer comes from. A field in the ERP. The spec sheet. A person to ask. You can also tell it where not to look.
Past jobs become tests.
SMEme asks for past cases with known outcomes, including one that went the less obvious way. It turns those into test cases.
When you choose Deploy, SMEme runs the tests first. If any of them comes out different than you said, it stops. When you edit the tree, you redeploy and SMEme runs the tests again. Until you do, the AI keeps running the last version you approved.
Then you test it on a job it hasn’t seen, in a fresh chat, without telling it the answer.
What this really shows: the software isn’t the hard part. Getting the knowledge out of someone’s head and agreeing on it is. SMEme makes you do that step on purpose.
Part 3 covers what you can ask a running tree, and who owns it.
Tools used:
- Claude (read the SMEme docs, drafted this article)
- SMEme (the tool under discussion)


Comments