Select all components that are required to build a minimal end to end RAG system (the absolute basics, not nice to haves like rerankers or eval harnesses).
The correct answers are 0, 1, 2, and 5: embedding model, vector store, generator LLM, and chunker. Together they form the four moving parts of a minimal **RAG** system.
Here is the walkthrough. Start with a folder of documents. First, the chunker splits each document into smaller pieces (typically 200 to 800 tokens). Second, the embedding model turns each chunk into a vector. Third, the vector store keeps those vectors and lets you find the closest ones to a new query vector. Fourth, when a user asks a question, you embed the question, fetch the top-k closest chunks from the store, and pass them to the generator LLM along with the question. The LLM writes the answer.
The two distractors are option 3 (fine-tuning) and option 4 (reranker). A reranker improves quality but is not required for a minimal system. Fine-tuning is not part of RAG at all; RAG is specifically the pattern where the model stays frozen and knowledge comes from retrieval. If you fine-tune the generator on your docs, you have left the pattern.
The minimal **RAG** architecture has exactly four required components, and naming them cleanly is what the question is testing.
First, a chunker at indexing time that splits source documents into retrieval-sized pieces, typically 200 to 800 tokens. Without chunking, you would embed whole documents as single blurry vectors and lose all retrieval precision.
Second, an **embedding model** like `text-embedding-3-large`, Voyage Embed v3, or Cohere Embed v4 that converts both chunks (at indexing time) and queries (at search time) into vectors. The same model must be used on both sides.
Third, a vector index like Pinecone, Qdrant, pgvector, Weaviate, or Chroma that stores the chunk vectors and supports approximate nearest neighbor search at query time. Without ANN, retrieval would be O(N) and unusable at scale.
Fourth, a **generator LLM** like Claude Opus 4.7, GPT-5.5, Gemini 3.1, or Llama 4 that reads the retrieved chunks plus the user query and synthesizes an answer.
The distractors test for two common confusions. A reranker (option 4) is a strong quality lift that production systems add on top, but the minimum works without one. Fine-tuning the generator (option 3) is the more telling trap: it is explicitly outside the RAG pattern. RAG is built around a frozen model whose knowledge is supplied by retrieval, not weight updates. Selecting option 3 signals confusion about what RAG actually is.
Quick reference
TL;DR
Minimal RAG has four required parts: chunker, embedding model, vector index, and generator LLM. Rerankers and fine-tuning are not on that list; one is a quality lift, the other leaves the pattern.
Picture the smallest RAG system you can build. You have a folder of documents and a user with a question. To go from one to the other, you need four things. Something to chop the documents into bite size pieces, so retrieval can be precise. Something to turn each piece (and the question) into a number coordinate, so 'similar meaning' becomes a measurable distance. Something to store those coordinates and find the closest ones fast. And something to read the closest pieces and write an answer for the user. Take any one of those four away and the system stops working. Rerankers, fine-tunes, eval harnesses, and clever query rewriters are all good ideas later, but none of them are in the minimum kit. And fine-tuning is not even part of the pattern: it changes the model, which RAG is specifically designed to avoid.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
Corpus is small enough to fit in the system prompt Skip embeddings, vector index, and retrieval entirely; the model reads the whole corpus directly, removing three of the four RAG components.
You need a style or format change rather than fresh facts Fine-tuning the generator adapts voice and structure; RAG adds evidence but does not change style. Pick the mechanism that matches the goal.
Queries are pure exact match lookups by identifier A relational index or BM25 keyword search retrieves exact tokens reliably; the embedding plus vector index stack adds latency without precision.
Quick scan · 2–4 min
Everything important, quickly.
3 min: name the four components in order (chunker, embedder, index, generator), walk through the index-time and query-time flows, then call out the reranker and fine-tuning distractors and why each is wrong.
Real-world examples
Real products, models, and research that use this idea.
1LangChain's basic RAG template shipping exactly these four parts: TextSplitter, OpenAIEmbeddings, Chroma vector store, and a ChatOpenAI generator.
2LlamaIndex's quickstart pipeline using SimpleNodeParser, OpenAI embeddings, an in-process vector store, and Claude Opus 4.7 as the generator.
3Pinecone's reference RAG architecture with a chunking utility, OpenAI text-embedding-3-large, Pinecone serverless, and GPT-5.5 for generation.
4pgvector + Postgres-based RAG inside many Django and Rails apps, with manual chunking, OpenAI embeddings, pgvector storage, and any frontier model for generation.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QHow would you extend the minimal four-part system to make it production-ready?
Sign in or go Pro to see the approach for this follow-up.
QWhy is fine-tuning the generator considered to leave the RAG pattern?
Sign in or go Pro to see the approach for this follow-up.
QIf you could only add one optional component on top of the minimum, which would it be and why?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Listing a reranker or a fine-tuned model as part of the minimum architecture. Rerankers are a quality lift on top; fine-tuning leaves the RAG pattern entirely because RAG is built around a frozen generator.