Intelligent Coaching Application

Abstract

This thesis presents the design and implementation of a mobile fitness coaching application powered by an AI agent. The proposed system tackles the fragmentation of current fitness applications by integrating workout planning, exercise logging, nutrition and hydration track￾ing, progress visualization, AI-powered food image analysis, coach-athlete communication, and personalized guidance into a single unified platform. The mobile client was developed using React Native and Expo, while the backend was built with Bun and ElysiaJS. Data persistence is managed through PostgreSQL with Prisma ORM, enhanced by the pgvector extension to enable semantic search for intelligent recom￾mendations. The AI layer integrates Google Gemini for conversational coaching and food image analysis, alongside Ollama and the nomic-embed-text model to deliver retrieval￾augmented workout recommendations. The backend was deployed on Fly.io with monitoring and logging mechanisms to ensure operational reliability. The resulting application demonstrates how user context, structured fitness data, and AI agent workflows can be effectively combined to provide more adaptive coaching, improve tracking consistency, and support better long-term training and nutrition decisions.