Integration of a large language model as an agent of the ChronobioticsDB pharmacological database

Objective. To develop a methodologically rigorous approach for employing large language models as an interface for accessing data from pharmacological repositories (exemplified by ChronobioticsDB). Material and methods. A method for integrating a large language model as an intelligent access agent for the ChronobioticsDB pharmacological database was proposed. The approach is based on two-stage meta-prompting and does not require additional model training. Relational database data are pre-transformed into the structured text cards, which are used as a sole source of information for the model. A post-processing module is additionally implemented. This module provides control over format, language and response sources. Results. Original system was tested on a set of queries, including factual, comparative, and data-incomplete queries. In all cases, the agent generated correct responses based solely on ChronobioticsDB data without generating internal reasoning or unreliable information. Conclusion. A protocol and code for integrating the DeepSeek R1 (8B) agent into specialized pharmacology databases and clinical guidelines were developed. This approach can be considered a promising alternative to retraining models for clinical pharmacology and other disciplines.