This study presents a personalized portfolio optimization framework that integrates behavioural insights, non-quadratic preferences, and metaheuristic optimization. Moving beyond the classical mean-variance paradigm, we model investor preferences using three distinct utility functions—power, exponential, and logarithmic—each capturing different attitudes toward risk, skewness, and downside aversion. These utility functions reflect more realistic, behaviourally motivated decision patterns in financial settings. To solve the resulting non-convex optimization problems, we employ Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), two metaheuristic techniques well-suited for handling higher-moment objective functions. We evaluate and compare their performance across utility types using real-market return data. Notably, this is the first study to contrast PSO and GA under three structurally distinct utility functions in a real-market context, offering insights into how algorithmic behaviour aligns with varying investor archetypes. Robustness checks—including box plot analyses and Wilcoxon Signed-Rank Tests—confirm the statistical significance of performance differences. Across all utility types, PSO demonstrates consistently superior performance, with higher median utilities, lower variability, and fewer extreme outcomes. While GA occasionally yields slightly higher average utility values, PSO’s stability and robustness make it a more reliable choice for portfolio construction under behaviourally grounded preferences. These findings highlight the value of personalized optimization approaches in financial decision-making and offer actionable guidance for portfolio managers and fintech platforms aiming to tailor investment strategies to individual behavioural profiles.