A new framework has been introduced to improve Grounded Video Question Answering (GVQA) in preparation for the ICCV 2025 Perception Test Challenge. This approach aims to develop robust multimodal large language models capable of reasoning over video content while visually grounding their answers. A key feature of the framework is its ability to track referenced objects across time, enhancing spatio-temporal grounding in video analysis. By focusing on both the temporal and spatial aspects of video data, the framework seeks to address challenges in accurately pinpointing trigger moments relevant to questions posed about video content. This development represents a significant step toward more sophisticated multimodal understanding in AI systems. The framework's introduction and focus have been documented in recent research shared on arXiv under the computer vision category. These advancements align with ongoing efforts to enhance video-based question answering capabilities in large language models.