<p>Obesity, characterized by excess body fat, increases the risk of diabetes, cardiovascular disease, and musculoskeletal disorders. Traditional metrics such as Body Mass Index (BMI) often misrepresent body composition, necessitating more accurate predictive models. Machine learning (ML) offers promise but typically requires complex, expertise-driven workflows. This study presents a proof-of-concept multi-agent framework that leverages Large Language Models (LLMs) and Chain-of-Thought (CoT) reasoning to automate body fat percentage prediction and generate personalized health recommendations. Using a structured dataset comprising 252 samples with anthropometric features, the system employs a coordination layer to manage specialized agents responsible for data preprocessing, model training, evaluation, and recommendation generation. CoT reasoning enables adaptive task optimization by dynamically adjusting workflows—such as imputation and hyperparameter tuning—based on dataset characteristics, thereby reducing manual intervention. Model interpretability is achieved through feature attribution methods, while an interactive user interface facilitates visualization and real-time feedback. The CoT-enhanced framework outperforms both static-prompt multi-agent systems (MSE: 6.78, R²: 0.84) and manual baselines (MSE: 18.19, R²: 0.748), with XGBoost achieving the best results (MSE: 4.12, R²: 0.92). Personalized outputs—including diet, exercise, and lifestyle guidance—are generated for each user. The proposed system reduces development time (≈ 95%) and demonstrates potential for rapid, scalable and interpretable healthcare AI prototyping.</p>