RAG Engineering Mastery6 / 10
Prompting the Generator — Grounding & Citations
Great retrieval is wasted if the model ignores it or can't point to its sources. Grounding is a prompt-design discipline, not an afterthought.
You retrieved the right chunks. Now the model has to use them — and only them. Grounding is the prompt discipline that turns retrieved context into a trustworthy answer.
The three rules of a grounded prompt
- Answer only from context. State it explicitly: "Use only the provided sources. If they don't contain the answer, say so."
- Cite by id. Give each chunk an id and require inline citations like
[3]. Citations make answers auditable and build user trust. - Permit "I don't know." An honest gap beats a confident fabrication. Reward abstention in the instruction.
Structure the context
SOURCES:
[1] (title, url) … chunk text …
[2] (title, url) … chunk text …
QUESTION: …
Answer using only the sources above. Cite as [n]. If the sources
do not answer the question, say you don't know.
Order matters: put the strongest re-ranked chunks first, and keep the source block visually distinct from the instruction.
Series — RAG Engineering Mastery
- Part 01Why Naive RAG Fails in ProductionThe 50-line vector-search demo that wows in a notebook falls apart the moment real users ask real questions. Here is why — and the map out.
- Part 02Chunking — The Decision That Sets Your CeilingYou can't retrieve what you chunked badly. Chunking is the most under-rated lever in RAG — and the cheapest to get right.
- Part 03Embeddings & Vector Stores 101An embedding turns meaning into geometry. A vector store makes that geometry searchable in milliseconds. Get both right and retrieval gets easy.
- Part 04Hybrid Retrieval — Keyword + VectorVector search understands meaning but fumbles exact terms, IDs, and rare words. Keyword search nails those and misses paraphrase. Use both.
- Part 05Re-Ranking — The Cheap Quality WinRetrieval gets you 30 plausible chunks. A re-ranker reads them against the actual question and floats the truly relevant few to the top.
- Part 06Prompting the Generator — Grounding & Citations — you are hereGreat retrieval is wasted if the model ignores it or can't point to its sources. Grounding is a prompt-design discipline, not an afterthought.
- Part 07Evaluation — You Can't Improve What You Don't MeasureWithout an eval set, every RAG change is a vibe. With one, you tune chunking, retrieval and prompts with a number that tells you if you helped or hurt.
- Part 08Handling Hallucinations & GuardrailsWhen retrieval comes up empty, a helpful model invents. Guardrails turn 'confidently wrong' into 'honestly unsure' — the difference users actually trust.
- Part 09Cost & Latency DisciplineA RAG query touches embeddings, a vector DB, a re-ranker and an LLM. Each adds milliseconds and cents. At scale, discipline here is the difference between a margin and a bonfire.
- Part 10The Production RAG Reference ArchitectureEvery piece, assembled: ingestion, hybrid retrieval, re-ranking, grounded generation, guardrails, eval and caching — the blueprint you can ship.