| Abstract [eng] |
This project analyses the usage of the DSS based on large language models in the context of the food product development process in the food industry. As an object of analysis, a yeast-raised product "Lemon Curd Poppy Seed Loaf" is chosen, the present recipe and production of which does not provide the required level of quality stability during 6 months of frozen storage. The aim of the research is to implement an LLM-based DSS that would provide quality improvement and increase efficiency of food product development in the industry. In the theoretical part of the paper, Food Quality 4.0 and Food Processing 4.0 trends, application of LLMs for decision support, RAG and HITL approaches, and factors of deterioration of quality of frozen yeast-raised products: water migration, decrease in yeast activity, weakening of gluten structure, and formation of ice crystals are investigated. On the basis of theoretical analysis, a structure for creation of LLM-DSS is developed, considering recipe, process parameters, raw materials requirement, and quality criteria. In practice, the suggested system is used in three different scenarios - the control recipe (K0), modified recipe (S1), in which part of sugar is replaced with glucose syrup and inulin, while the amount of water is decreased, and modified recipe and process (S2), where part of the flour is replaced with wheat malt, the "CO2MMITTED BRIOCHE FREE 5%" improver content is increased, and proofing temperature is reduced. After four months of freezing, a sensory analysis and an ImageJ analysis of crumb structure were conducted. S1 received the highest score in sensory testing - 8,0 points, while S2 had the smallest area of structural objects in ImageJ – 56,4% relative to K0. Economical and managerial evaluation demonstrated that usage of LLM-DSS decreased the size of the decision search space up to 66,7%. According to this evaluation, recipe modification S1 is preferred as the most applicable option for further implementation due to its quality improvement and lack of necessity for process modifications. |
