In this chapter
brAIn is organized knowledge from a particular domain. For us, that includes civil engineering: PDFs, articles, laws, historical studies, and expert knowledge. I want to use it for a specific question without searching every material again from the beginning.
brAIn could be a separate product. That knowledge has value even without connecting it to an application. We took a further step and connected it with the code model. We call that connection MiddleBrain.
Knowledge needs a source
A convincing-sounding answer is not enough. We need material from which the rule, its scope, and when it applies follow. A historical study may explain an old decision but not current requirements.
When describing a rule, record where it comes from, what it concerns, and when to ask an expert. If sources conflict or omit something, the agent should show that gap, not invent the missing rule.
How we use brAIn for a new feature
We begin with: what must this feature know and take into account to do its job correctly? We use domain knowledge and, after connecting it to code, check what data and functions already exist and what is still missing.
Knowledge must become something concrete: needed data, operating rules, exceptions, and a way to check the result. That gives a person who is not a civil-engineering expert a basis for analysis and delivery, and gives the user a feature that includes that knowledge in its operation.
When someone reports: “That is not how it works”
We also use brAIn to investigate defects. We check what domain materials say, compare them with application behaviour, and consult the conclusion with an expert as a safeguard. An AI answer alone does not close the report.
For such a report, record side by side: the user case, the rule with its source, the application result, and the agreement with the expert. Then it is clear whether to correct code, complete knowledge, or explain feature use.
How we organize materials
Open Knowledge Format (OKF), developed in the GoogleCloudPlatform repository, organizes knowledge in Markdown files with YAML metadata — a description of the source and context stored alongside content. It is a way to organize knowledge, not ready-made domain knowledge.
The file format alone does not confirm that a rule is correct. Sources and their verification are needed. I describe the JSON code-dependency model separately under middleware.
Product creator
Paweł Ochmann