Nutrition Recommender and Adjustment System by Fine-tuning LLMs

The advent of large models(LLMs) has revolutionized the landscape of artificial intelligence, particularly influencing sectors like healthcare and nutrition. These models possess an extraordinary capacity for understanding and generating natural language, and their ability to generalize knowledge across various domains becomes even more powerful when fine-tuned for specific tasks. Leveraging transfer learning, LLMs can apply knowledge from one context to expedite learning in another. The major problem with available Systems is, not providing items which are not accessible or culturally appropriate for users by fine tuning Gemini model on finding region of user by their diet and limiting it to suggest only food items that are from same region solves this problem. By an available dataset from Kaggle which has data of nutrient content of some specific item and using Schofield Equation to calculate the max and min amounts of macronutrients that the user needs per day, calculated deficient and excess nutrient by basing this the LLM(Gemini) will return appropriate diet by changing, adding or removing the user’s diet. As a result, finetuned LLM(Gemini) has successfully identified and suggested food items from the same region. In this paper, we discussed methods to optimize the nutrition recommender and adjustment system based on user preferences, highlighting the fine-tuning of large language models (LLMs).