Context-based prompt generation for automated translations between natural language and query language

A system comprising:

• a pool of shots, each shot in the pool including a natural language query component and a corresponding database translation component that translates the natural language query component into a query language used by a particular database;
• a trained semantic similarity model that generates vectorized shots by vectorizing information within each of the shots into a common vector space;
• a query translator stored in memory and executable to:
• receive, from a user interface, a natural language query requesting information stored in the particular database;
• generate a request vector by vectorizing the natural language query into the common vector space;
• identify a similar subset of the vectorized shots that satisfy a similarity metric when compared to the request vector;
• generate a large language model (LLM) prompt that includes:
• a group of the shots from the pool corresponding to the similar subset of the vectorized natural shots;
• and
• a request to translate the natural language query into the query language used by the particular database;
• provide the LLM prompt as input to an LLM;
• and
• receive, from the LLM, a translation of the natural language query into the query language used by the particular database.

One or more tangible computer-readable storage media encoding processor-executable instructions for executing a computer process comprising:

• obtaining a pool of shots that each include a natural language query component and a corresponding database translation component that translates the natural language query component into a query language used by a particular database;
• generating vectorized shots by vectorizing, with a trained semantic model, information within each of the shots within a common vector space;
• receiving, from a user interface, a natural language query requesting information stored within the particular database;
• vectorizing, by the trained semantic model, the natural language query to generate a request vector defined within the common vector space;
• identifying a similar subset of the vectorized shots that satisfy a similarity metric when compared to the request vector;
• generating a large language model (LLM) prompt that includes:
• a group of the shots from the pool corresponding to the similar subset of the vectorized shots;
• and
• a request to translate the natural language query into the query language used by the particular database;
• providing the LLM prompt as input to an LLM;
• and
• receiving, from the LLM, a translation of the natural language query in the query language used by the particular database.