The rapid expansion of digital content, particularly multimedia, necessitates efficient and accurate information retrieval mechanisms. Traditional search engines primarily rely on keyword-based indexing techniques, often leading to suboptimal search results due to a lack of contextual understanding. This research explores the integration of Retrieval-Augmented Generation (RAG) for enhancing transcript-based search functionality, replacing traditional indexing methods such as Elasticsearch with a dynamic AI-driven pipeline. Utilizing LangChain, LangGraph, and ChromaDB, we develop a scalable and context-aware retrieval system that processes video transcripts and applies the GPT-3.5-Turbo model to refine search queries and generate insightful responses. The proposed system leverages vector embeddings to index and retrieve relevant transcript segments, ensuring high semantic accuracy in search results. Our experimental evaluation compares the effectiveness of this approach against conventional search methods, measuring accuracy, retrieval relevance, and user experience. The results demonstrate the superiority of RAG-based retrieval in delivering precise, contextually relevant information, paving the way for advanced AI-driven search systems.