Worked on the nvidia-holoscan/holohub repository to enhance the robustness of data ingestion for the EndoNeRF loader, focusing on surgical_scene_recon datasets. Addressed a critical bug by aligning image, depth, and mask frame counts to the minimum available frames, ensuring consistent data processing and validation. Implemented truncation logic to handle mismatched frame counts, added warning emissions for truncation events, and introduced clear error handling when no usable frames remain. Utilized Python for debugging and error handling, improving the reliability of model training and evaluation pipelines. These changes stabilized data loading and enhanced observability for downstream debugging and monitoring workflows.