Checking for potential source code having an issue by augmenting a natural language prompt with source-code-describing content that is similar to the issue

A computer-program product comprising one or more non-transitory machine-readable storage media, including stored instructions configured to cause a computing system to perform a set of actions including:

• generating a plurality of vector embeddings representing a plurality of sets of content describing functionality of a plurality of sets of code;
• accessing natural language describing an issue detected in a set of code;
• generating an issue vector embedding representing the issue based at least in part on the natural language;
• determining a plurality of distances between the plurality of vector embeddings and the issue vector embedding;
• based at least in part on the plurality of distances, selecting, to support communication with a large language model, a subset of the plurality of sets of content describing functionality of a subset of the plurality of sets of code;
• generating a prompt for execution at least in part by the large language model, wherein the prompt comprises:
• at least some natural language describing the issue, and
• at least some content from each set of content in the subset of the plurality of sets of content;
• prompting the large language model with the prompt;
• accessing a result based at least in part on the execution;
• and
• storing a new association between a particular set of code and the issue based at least in part on the result.

A system comprising:

• one or more processors;
• one or more non-transitory computer-readable media storing instructions, which, when executed by the system, cause the system to perform a set of actions including:
• generating a plurality of vector embeddings representing a plurality of sets of content describing functionality of a plurality of sets of code;
• accessing natural language describing an issue detected in a set of code;
• generating an issue vector embedding representing the issue based at least in part on the natural language;
• determining a plurality of distances between the plurality of vector embeddings and the issue vector embedding;
• based at least in part on the plurality of distances, selecting, to support communication with a large language model, a subset of the plurality of sets of content describing functionality of a subset of the plurality of sets of code;
• generating a prompt for execution at least in part by the large language model, wherein the prompt comprises:
• at least some natural language describing the issue, and
• at least some content from each set of content in the subset of the plurality of sets of content;
• prompting the large language model with the prompt;
• accessing a result based at least in part on the execution;
• and
• storing a new association between a particular set of code and the issue based at least in part on the result.