Memory-Augmented Vision-Language Agents for Persistent and Semantically Consistent Object Captioning | Signal Canvas | ScienceToStartup