In short
A RAG (retrieval-augmented generation) engineer builds the systems that fetch relevant documents and data and feed them to a language model so its answers are grounded and current. As of 19 September 2026, AgenticCareers.co tracks 33 distinct RAG and search roles at 24 companies, mostly senior, mid and staff level. Too few listings disclose pay (2 of 33) to publish a reliable salary median.
Language models only know what they were trained on. Retrieval-augmented generation, or RAG, fixes that by fetching relevant documents at question time and handing them to the model along with the question. A RAG engineer builds and tunes that retrieval layer: what gets indexed, how it is split and embedded, how the right passages are found, and how the system proves its answers are grounded. For the architecture itself, our practical guide to RAG pipelines goes step by step.
What a RAG engineer does day to day
- Ingestion and chunking. Turning messy sources (PDFs, wikis, tickets, databases) into clean, well-sized passages with the metadata needed to filter them.
- Embeddings and indexes. Choosing embedding models and a vector or hybrid index, and keeping it fresh as sources change.
- Retrieval quality. Combining keyword and vector search, re-ranking, and query rewriting so the right passage lands in the top few results.
- Evaluation. Measuring whether answers are correct and supported by the retrieved text, and catching regressions when anything in the pipeline changes.
- Serving and cost. Keeping latency and token spend under control in production.
RAG engineer vs ML engineer vs search engineer
The three roles overlap, and job titles blur them, so it helps to know the difference in emphasis:
- ML engineers train, fine-tune and deploy models. Their focus is the model itself.
- Search engineers build ranking and relevance systems, traditionally for users typing queries into a search box.
- RAG engineers sit between the two: they use search and ranking techniques, but the consumer of the results is a language model, not a person. That changes what "good" means. The model needs a small number of highly relevant, well-formed passages with the right context, and the final measure is whether the generated answer is correct and supported.
In practice, many RAG openings are posted as applied AI, LLM or search roles. Of the 33 roles we track, 22 sit in our RAG and search category, 6 in LLM engineering and 3 in AI agent engineering.
Who is hiring RAG engineers
As of 19 September 2026 we track 33 distinct RAG and search roles at 24 companies. Few use the literal title "RAG Engineer". Look for titles like Staff Software Engineer, AI Search, Senior Staff Applied AI Engineer, Context Retrieval, Senior AI Data Engineer, Data Products and RAG Foundations or Lead Software Engineer (Python, Vector Databases). Databricks has the most openings (4), followed by Anthropic, MongoDB and Wells Fargo with 2 each, which shows the demand spans AI labs, data platforms and large enterprises. Browse the full set on the RAG and Search role page.
Seniority
The roles skew experienced: 11 senior, 10 mid-level, 9 staff and 3 lead. There is no junior opening in the set right now. Engineers usually move into RAG work from backend, search or data engineering, where indexing and query performance are already familiar.
Skills employers ask for
Across the postings, the recurring requirements are strong Python, experience with vector databases and search engines, embedding models, evaluation of retrieval and answer quality, and cloud deployment. Several enterprise roles ask specifically for Azure-based generative AI work. The rarest and most valuable skill is evaluation: teams can build a pipeline in a week, but proving it retrieves the right thing takes discipline.
What interviews tend to test
Expect to talk through a retrieval system you built and defend its choices: how you chunked documents and why, when keyword search beats embeddings (exact names, codes and numbers), how you would add re-ranking, and how you measured retrieval and answer quality. A common exercise is diagnosing a pipeline that returns plausible but wrong answers; the strongest candidates check retrieval first, because most wrong answers start with the wrong passage.
What about pay?
Only 2 of the 33 listings disclose a salary, below the 30 we require before publishing a median, so we do not give one here. Related roles with more disclosed data are on our salary data page.
How to get into RAG engineering
Build one retrieval system end to end on real, messy documents, then measure it: report how often the correct passage appears in the top results and how often answers are supported by the retrieved text. That evaluation evidence is what separates candidates in interviews, because it is the part most teams struggle with.
Frequently asked
What does RAG stand for?
Retrieval-augmented generation: fetching relevant documents at question time and giving them to a language model so its answer is grounded in current, specific information.
How many companies are hiring RAG engineers?
As of 19 September 2026, AgenticCareers.co tracks 33 distinct RAG and search roles at 24 companies, with Databricks hiring the most (4).
Is RAG engineer an entry-level job?
Not usually. Current openings are senior (11), mid-level (10), staff (9) and lead (3), with no junior roles. Most people move in from backend, search or data engineering.
What skills does a RAG engineer need?
Python, vector databases and search engines, embedding models, retrieval and answer-quality evaluation, and cloud deployment. Evaluation skill is the scarcest.