Аңдатпа
Dietary habits play a critical role in affecting health outcomes. Present applications and algorithms developed to provide recommendations to change dietary habits are often focused on Western or globally popular cuisines, neglecting the unique culinary landscapes of other regions. Central Asia, unlike Western diets, emphasizes culturally significant dishes that incorporate local ingredients and traditions that require specialized consideration for dietary recommendations. Our study addresses this gap by developing a dietary recommendation system tailored to Central Asian cuisine. With the help of a computer vision model that was specifically trained on a Central Asian Food Dataset (CAFD), our algorithm accurately identifies the food items on the image. To enable the practical integration of Large Language Models (LLMs) in personalized dietary planning, we developed an algorithm designed to construct dynamic prompts that incorporate a combination of simulated patient profiles and a synthesized dataset containing textual descriptions of a diverse selection of Central Asian dishes. This algorithm tailors meal suggestions by taking into account individual health parameters, including specific health conditions, medical history, and lifestyle factors. It further enhances the relevance of its recommendations by emphasizing the nutritional and cultural significance of Central Asian cuisine. Preliminary evaluations demonstrate that the system offers a scalable framework for personalized meal planning, particularly for Central Asian countries. This work proposes a prototype for a digital nutrition assistant, leveraging both image and text data and carried out the experiments using different LLMs (i.e. ChatGPT, Gemini, LLama).