In-context learning with templates for large language model generation of customized emails

A system for using a trained large language model (LLM) to generate a customized email, the system comprising:

• memory;
• and
• an LLM context confiner stored in the memory that:
• receives, from an email application, an email template and a recipient identifier that identifies a recipient of the customized email;
• mines contextual data associated with the recipient identifier, at least some of the contextual data pertaining to past communications between a sender of the customized email and a recipient identified by the recipient identifier;
• generates a large language model (LLM) prompt that includes based on the email template and the contextual data, the LLM prompt requesting as output the customized email;
• receives the customized email as an output from the LLM in response to providing the LLM with the LLM prompt;
• and
• returns the customized email to the email application.

One or more computer-readable storage media encoding processor-executable instructions for executing a computer process that uses a trained large language model (LLM) to generate a customized email according to a methodology that reduces a risk of the customized email including hallucinated facts and personal identity information (PII), the computer process comprising:

• receiving a selection of a email template and a recipient identifier that identifies a recipient of the customized email;
• mining contextual data associated with the recipient identifier, at least some of the contextual data pertaining to past communications between a sender of the customized email and a recipient identified by the recipient identifier;
• executing scoring logic to identify relevant contextual information from the contextual data associated with the recipient identifier, wherein the relevant contextual information comprises information that is determined to be relevant to at least one of the email template and a sender of the customized email;
• generating a large language model (LLM) prompt that includes the email template and the relevant contextual information, the LLM prompt requesting as output the customized email;
• and
• presenting, on a user display, the customized email within an email composition window.