RAG

Retrieval-Augmented Generation (RAG) is an architectural pattern that enhances large language model outputs by retrieving relevant documents from external knowledge bases before generation, addressing limitations around hallucination and outdated training data. Job listings requiring RAG expertise typically come from companies building AI applications over proprietary knowledge—customer support systems, internal documentation search, legal research tools, or domain-specific question answering where accuracy and attribution matter more than creative generation. ML engineers and AI application developers are expected to implement document chunking strategies, choose appropriate embedding models and vector databases, design retrieval ranking algorithms, and optimize the trade-off between context window usage and relevance. The pattern's effectiveness depends on data quality, chunking granularity, and prompt engineering to ensure models properly utilize retrieved context. Roles often involve building evaluation frameworks to measure retrieval quality separately from generation quality, implementing hybrid search combining semantic and keyword matching, and handling multi-turn conversations where context accumulates. Companies investing in RAG typically have large document repositories or require models to answer questions about information beyond their training cutoff.