Abstract: Apparatus and methods are disclosed for implementing a copilot as a network of microservices including specialized large language models (LLMs) or other trained machine learning (ML) tools. The microservice network architecture supports flexible, customizable, or dynamically determinable dataflow. Compared to much larger competing LLMs, comparable or superior performance is achieved for certain tasks, while significantly reducing hardware requirements and computation time. Disclosed examples incorporate microservices for expansion, retrieval, embedding, and evaluation, in addition to one or more core microservices. Optionally, intermodal I/O, multiple data repositories, competency qualification, or human feedback can be supported. Multiple core microservices can support varying client authorizations or cognitive functions. The disclosed architecture supports any major LLM use case and can be deployed on a single compute node with a single GPU.

Abstract: Computerized methods are disclosed for automated question generation from source documents through natural language processing, for applications including training and testing. Interleaved selection and transformation phases employ combined semantic-syntactic analysis to progressively refine natural input text into a high density of text fragments having high content value. Non-local semantic content and attributes such as emphasis attributes can be attached to the text fragments. The text fragments are reverse parsed by matching against a precomputed library of combined semantic-syntactic patterns. Once the patterns of each fragment are determined, transformation of fragments into question-answer pairs is performed using question selectors and answer selectors tailored to each pattern. Methods for constructing distractors, both internal and external, are also disclosed. The ecosystem of machine learning components, ontology resources, and process improvement are also described.

Abstract: Computerized methods are disclosed for automated question generation from source documents through natural language processing, for applications including training and testing. Interleaved selection and transformation phases employ combined semantic-syntactic analysis to progressively refine natural input text into a high density of text fragments having high content value. Non-local semantic content and attributes such as emphasis attributes can be attached to the text fragments. The text fragments are reverse parsed by matching against a precomputed library of combined semantic-syntactic patterns. Once the patterns of each fragment are determined, transformation of fragments into question-answer pairs is performed using question selectors and answer selectors tailored to each pattern. Methods for constructing distractors, both internal and external, are also disclosed. The ecosystem of machine learning components, ontology resources, and process improvement are also described.