Large-scale video archives, such as the NHK Archives with over ten million assets, present significant content retrieval challenges. Traditional reliance on labor-intensive, costly, and often incomplete manual metadata severely restricts discoverability. While AI offers solutions, cloud-based AI introduces critical data security concerns for sensitive visual assets. This necessitates innovative AI-driven search solutions that are powerful and intuitive, yet prioritize data privacy, cost-effectiveness, and operational efficiency for extensive archives. This report proposes a novel methodology to address these multifaceted challenges.