Integration of large language models into chatbot-based customer request processing systems.

The advancement of artificial intelligence has facilitated the emergence of a new generation of tools for automating customer inquiry processing, particularly through the widespread implementation of large language models (LLMs). These models have unlocked new possibilities in the development of dialog systems. The integration of LLMs into chatbots has become a key factor in transforming customer service, ensuring a high level of personalization, contextual relevance, and linguistic flexibility in user interactions. The aim of this article is to analyze the role of large language models in the development of intelligent customer service systems and to design architectural and algorithmic solutions for their effective integration into chatbot platforms. The study adopts a systemic approach to assess the functional potential of LLMs in multichannel communication, applies comparative analysis to explore current approaches to dialog agent implementation, and employs structural-functional modeling to build an integration architecture for LLMs in client services. It has been established that models such as GPT, LLaMA, Claude, and other transformer-based architectures enable not only the automation of routine inquiries but also the handling of complex requests that require contextual understanding, sentiment analysis, and user intent recognition. Key advantages include reducing the workload of contact center operators, increasing system response speed, adaptive learning based on inquiry history, and generating linguistically relevant replies. At the same time, the study identified a range of challenges associated with integrating LLMs into business processes. These include the complexity of customizing models to domain-specific contexts, risks of generating inaccurate or ethically questionable content, high computational costs of inference, and the necessity to ensure secure processing of personal data. A conceptual integration model is proposed, comprising three core levels: (1) the input collection layer that gathers customer inquiries from various channels (messaging apps, web interfaces, mobile applications); (2) the processing layer based on LLMs, including preliminary sentiment and thematic analysis; and (3) the decision-making and response generation layer, with escalation to a human operator in the case of complex or critical scenarios. In conclusion, the article outlines recommendations for the effective implementation of LLMs in customer service systems: (1) selecting a hybrid architecture combining rule-based and generative components; (2) applying fine-tuning or Retrieval-Augmented Generation (RAG) approaches for domain adaptation; (3) maintaining continuous monitoring of generated responses to ensure quality, ethical compliance, and contextual relevance; (4) leveraging API gateways and edge computing to reduce response latency. Future research perspectives include enhancing models for multilingual support, integration with CRM/ERP systems, developing self-learning dialog agents, and establishing metrics to evaluate the effectiveness of generative responses in real-world environments