In this paper, we report three upgrades to our Interactive Video Search system. (i) A unified vector-database backend enables hybrid dense-sparse retrieval with in-engine filtering and late-fusion re-ranking, improving responsiveness under mixed embedding search and text-based search. (ii) A self-similarity-matrix (SSM) shot segmentation pipeline adapts to both short and long videos, yielding more coherent units for retrieval and browsing. (iii) We generalize the VIREO grid-sketch from colors to an object-sketch: object-by-cell atoms are precomputed offline and combined interactively with the same recommendation shading as the original interface. Across the Vimeo Creative Commons Collection (V3C), Marine Video Kit (MVK), and Laparoscopic Hysterectomies (LHE75) datasets, these updates increase search quality while preserving the system’s interactive workflow.