RAG & Search jobs

Retrieval roles own the pipeline that decides what the model gets to know: chunking, embeddings, hybrid search, rerankers, and the eternal fight against confidently wrong answers sourced from the wrong document. Everyone built a naive RAG demo in 2023; these jobs exist because the naive version plateaus at “mostly right”, and companies need better.

Vector stores like pgvector and Qdrant share racks with Elasticsearch, and old-school information-retrieval instincts — BM25 first, measure everything — age remarkably well. Eval sets that score faithfulness as well as relevance separate serious teams from demos. Enterprises with big document estates hire this constantly, and the work is almost always remote-compatible.