Systems and methods for zero-shot, fast-generation and implementation of an intelligent virtual dialogue agent using one or more pre-trained machine learning-based language models and a response corpus

Abstract

System and method of building a zero-shot training, machine learning-based virtual dialogue agent includes identifying a response corpus comprising a plurality of distinct response samples; providing the response corpus to a virtual agent generator; generating, by one or more pre-trained machine learning language models, a distinct embeddings inference for each of the plurality of distinct response samples; forming an embeddings-based architecture for response generation based on the distinct embeddings inference for each of the plurality of distinct response samples, wherein the embeddings-based architecture includes a mapping of the distinct embeddings inference for each of the plurality of distinct response samples to an n-dimensional space; instantiating a virtual dialogue agent based on receiving user stimuli; and computing, using the embeddings-based architecture, a response inference to the user stimuli, wherein the computed response inference is based on identifying one distinct embeddings inference of a distinct response sample of the response corpus.