Video content is growing exponentially, making effective search within videos a crucial challenge. Traditional video search engines largely rely on metadata (titles, tags, descriptions) and transcripts, which often fail to capture the rich visual and auditory content of videos​. This paper proposes an AI-based in-video search tool that indexes actual video content to enable granular search by scene, object, action, face, and speech. We integrate multiple deep learning models in a unified pipeline, object and face recognition to identify visual entities, action recognition to detect activities, and speech-to-text to transcribe spoken dialogue. The system automatically generates timestamped metadata and summaries for video segments, allowing content-based queries (“Robin Williams eating food scene”) that go beyond traditional metadata search. We evaluate the accuracy of each component and compare the overall system against existing solutions including text-metadata search and other similar research topics. Experimental results show that our content-based approach significantly improves recall of relevant moments in videos, demonstrating the advantages of AI-driven video search. We discuss challenges such as processing cost, imperfect AI-predictions, and suggest future improvements like expanded genre classification and multimodal semantic querying. The proposed system illustrates a step towards more intelligent video search engines that understand video content in depth.