Memory management system for appointment scheduling

A system for appointment scheduling, comprising:

• a processor;
• and
• a memory storing instructions that, when executed by the processor, cause the system to:
• generate one or more transcriptions from an audio stream of a live conversation;
• store a transcription of the one or more transcriptions in the memory when a relevance score of the transcription exceeds a threshold;
• delay invocation of a natural-language-understanding (NLU) model until a predetermined quantity of transcriptions is stored in the memory;
• upon the predetermined quantity of transcriptions being stored, invoke the NLU model on the stored transcriptions to produce a structured intent representation identifying an appointment-booking intent;
• and
• book an appointment at an external calendar service based on the structured intent representation,
• wherein delaying the invocation of the NLU model until the predetermined quantity of transcriptions is stored reduces transient memory usage and processor cycles during the live conversation.