Unhealthy dietary habits remain a global challenge, especially for individuals with specific health conditions. Existing mobile nutrition apps often prioritize calorie tracking over personalization or behavioral support. This study introduces HealKitchen, a persuasive, AI-driven mobile application grounded in behavioral science to promote healthier eating. HealKitchen personalizes recipe recommendations, identifies groceries through image recognition, and integrates persuasive features derived from five behavior-change theories. A mixed-method evaluation with 20 users showed high usability (mean = 4.6), usefulness (mean = 4.45), and aesthetics (mean = 4.49). Qualitative feedback emphasized the need for greater recipe diversity and richer multimedia. The findings demonstrate the feasibility of theory-based persuasive design for personalized nutrition support.