The paper presents a project of a functional architecture for an external control module integrated with a ChatGPT-class large language model (LLM). The module aims to support the stepwise development of subjectivity and functional elements of consciousness as components of artificial general intelligence (AGI). The approach is based on the conceptualization of consciousness as an emergent layer of cognitive organization, arising through gradual integration of system-level capabilities during interaction with the environment. The proposed architecture is structured into five layers: a consciousness shell, a subject kernel, a contextual field, a process manager, and an interface adapter. Each layer is supported by technical implementation using existing technologies, including LangChain, Redis, and PostgreSQL. The feasibility of the model is grounded in two key assumptions: (1) empirical evidence from neuroscience supports the graded formation of natural consciousness rather than its abrupt emergence, allowing the possibility of phased development in artificial systems; and (2) structural parallels exist between biological mechanisms of information encoding and the embedding-context structures in transformer-based architectures. Both systems employ context-dependent vector representations for the processing of semantically meaningful information. The proposed design provides a technical foundation for building engineered agents with functional features of consciousness. #COMESYSO1120