Building a Production-Ready RAG Pipeline: From Document Loading to Vector Retrieval