Semantic layer management techniques for code generation platforms

Abstract

Systems and methods are disclosed comprising techniques for maintaining a semantic layer in a code generation platform. The platform generates a first set of tokens by processing a natural language input. The platform generates a semantic layer knowledge data store comprising a table definition, a variable definition, and an aggregation context, The table definition comprises a database schema set, and the variable definition comprises a data object definition set. The aggregation context comprises a classification set generated based on the first set of tokens. The platform enables dynamically modifying the aggregation context based on a second set of tokens generated based on additional natural language input (e.g., a user query, query history), contextual information associated with the additional natural language input (e.g., a user role, persona, session context, and so forth), or both. At least a portion of the table definition and/or variable definition is dynamically indexed against the aggregation context.