Shvydko Decision Engine
This is the name we give to Shvydko's approach to decisions in code. You provide shared context, formulate questions and define criteria. Decision Engine returns a structured result: a choice, score or probability. You define the response format and next action before running the request.
Consider a support queue. A customer's inquiry describes a problem, but your system needs a concrete next task: route it to engineering, finance or logistics. Predefined options and criteria connect the text to an action in your workflow.
In Shvydko, this is a Choice question. Send the message with your routing rules and describe each allowed option. You can use the result in your own routing logic. The API itself does not move a ticket in your system; your integration performs that action.
Three response formats
The format depends on the next action. Choice fits category selection. Score uses an ordered scale with a description for each level. Noul estimates the probability of a statement. A single request can ask several questions about shared context.
For example, add questions about a support issue's impact and whether it requires human review. These are separate decisions: a technical category alone does not imply maximum urgency. Describe them separately so you can see which step produces an error.
- Choice: which team should handle the request?
- Score: how much does the issue interfere with placing an order?
- Noul: does the request need review under the stated rule?
From Playground to integration
Start with a ready-made scenario in the Playground. Read the context and criteria, replace them with your own example and run the request. Compare the response with the decision your rules lead you to expect. Also try a case with missing or conflicting information.
Once the format is clear, create a workspace API key and move the request into your server-side code. Keep the key out of the browser. Before automating, evaluate quality on a separate sample and decide which cases should go to a person.