Enhancing LLM Context-Awareness with RAG - the Nutrition Tracker Case

There are many nutrition trackers available on the market, but few leverage the full potential of artificial intelligence (AI) to deliver innovative user experiences. This thesis investigates novel approaches to enhance the functionality of nutrition trackers. The primary objective is to develop a system that extracts a macronutrient table from a user's input sentence without requiring manual searches for individual food items. To address this challenge, we explored several methods, including Named Entity Recognition (NER) and advanced techniques utilizing Large Language Models (LLMs). A Retrieval-Augmented Generation (RAG) framework was implemented, integrating various sources of information such as vector databases, web search tools, and model-embedded knowledge. By leveraging LLMs combined with precise prompting techniques, the proposed solutions demonstrated promising results, opening the way for more efficient and AI-driven nutrition tracking systems.