Interpretable Debiasing of Vision-Language Models for Social Fairness | Signal Canvas | ScienceToStartup