Formal Model and API Integration Method for Generative Neural Networks in a Web-Platform for Automated Text Processing

Abstract

The article presents a formal model for integrating generative neural network APIs into a corporate web platform for automated text processing. The proposed model is described by the tuple M = (U, C, R, P, F, A), where U is the set of users, C is the set of parameterizable prompts (contexts), R is the set of requests, P is the prompt parameterization function, F is the external API interaction function, and A is the audit and resource accounting function. The scientific novelty lies in the development of a formal model that ensures data isolation, centralized prompt management, and fault tolerance; in the proposal of a method for prompt parameterization based on contexts stored in the database; and in the creation of a robust API interaction algorithm that implements timeout handling, retries, and atomic result storage. An experimental study was conducted, in which the hypothesis of reduced text processing time using the proposed model was formulated and confirmed (p<0.001, effect size r=0.92). The developed platform has been deployed into industrial operation at the customer organization.