GenAI Ops · RAG & Prompting interview questions

1. A.input/output filters that block unsafe, off-topic or policy-violating LLM content
2. B.a parameterized prompt with slots for context and user input, versioned like code
3. C.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
4. D.fetching the k most similar chunks to the query to place in the prompt context

Answer + AI explanation with Pro

1. A.RAG
2. B.Hallucination
3. C.Top-k retrieval
4. D.Guardrails

Answer + AI explanation with Pro

1. A.RAG — a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
2. B.RAG — Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
3. C.RAG — fetching the k most similar chunks to the query to place in the prompt context
4. D.RAG — a parameterized prompt with slots for context and user input, versioned like code

Answer + AI explanation with Pro

1. A.a dense numeric vector representing text so that semantic similarity becomes vector distance
2. B.fetching the k most similar chunks to the query to place in the prompt context
3. C.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
4. D.an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce

Answer + AI explanation with Pro

1. A.Embedding
2. B.Prompt template
3. C.System prompt
4. D.Vector database

Answer + AI explanation with Pro

1. A.Embedding — splitting source documents into passages sized for embedding and retrieval
2. B.Embedding — fetching the k most similar chunks to the query to place in the prompt context
3. C.Embedding — a dense numeric vector representing text so that semantic similarity becomes vector distance
4. D.Embedding — input/output filters that block unsafe, off-topic or policy-violating LLM content

Answer + AI explanation with Pro

1. A.a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
2. B.the leading instruction that sets an LLM's role, constraints and tone for a session
3. C.input/output filters that block unsafe, off-topic or policy-violating LLM content
4. D.a store that indexes embeddings for fast approximate nearest-neighbor similarity search

Answer + AI explanation with Pro

1. A.Top-k retrieval
2. B.Vector database
3. C.Prompt template
4. D.Guardrails

Answer + AI explanation with Pro

1. A.Vector database — the leading instruction that sets an LLM's role, constraints and tone for a session
2. B.Vector database — a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
3. C.Vector database — a store that indexes embeddings for fast approximate nearest-neighbor similarity search
4. D.Vector database — fetching the k most similar chunks to the query to place in the prompt context

Answer + AI explanation with Pro

1. A.a parameterized prompt with slots for context and user input, versioned like code
2. B.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
3. C.fetching the k most similar chunks to the query to place in the prompt context
4. D.splitting source documents into passages sized for embedding and retrieval

Answer + AI explanation with Pro

1. A.Hallucination
2. B.Guardrails
3. C.System prompt
4. D.Chunking

Answer + AI explanation with Pro

1. A.Chunking — a dense numeric vector representing text so that semantic similarity becomes vector distance
2. B.Chunking — splitting source documents into passages sized for embedding and retrieval
3. C.Chunking — a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
4. D.Chunking — the leading instruction that sets an LLM's role, constraints and tone for a session

Answer + AI explanation with Pro

1. A.fetching the k most similar chunks to the query to place in the prompt context
2. B.a dense numeric vector representing text so that semantic similarity becomes vector distance
3. C.a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
4. D.input/output filters that block unsafe, off-topic or policy-violating LLM content

Answer + AI explanation with Pro

1. A.Top-k retrieval
2. B.Reranking
3. C.Prompt template
4. D.Chunking

Answer + AI explanation with Pro

1. A.Top-k retrieval — input/output filters that block unsafe, off-topic or policy-violating LLM content
2. B.Top-k retrieval — an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce
3. C.Top-k retrieval — fetching the k most similar chunks to the query to place in the prompt context
4. D.Top-k retrieval — the leading instruction that sets an LLM's role, constraints and tone for a session

Answer + AI explanation with Pro

1. A.an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce
2. B.a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
3. C.a store that indexes embeddings for fast approximate nearest-neighbor similarity search
4. D.a parameterized prompt with slots for context and user input, versioned like code

Answer + AI explanation with Pro

1. A.Prompt template
2. B.Reranking
3. C.Vector database
4. D.Embedding

Answer + AI explanation with Pro

1. A.Prompt template — fetching the k most similar chunks to the query to place in the prompt context
2. B.Prompt template — input/output filters that block unsafe, off-topic or policy-violating LLM content
3. C.Prompt template — a parameterized prompt with slots for context and user input, versioned like code
4. D.Prompt template — a second-stage model that reorders retrieved candidates by relevance before they enter the prompt

Answer + AI explanation with Pro

1. A.a dense numeric vector representing text so that semantic similarity becomes vector distance
2. B.a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
3. C.the leading instruction that sets an LLM's role, constraints and tone for a session
4. D.a parameterized prompt with slots for context and user input, versioned like code

Answer + AI explanation with Pro

1. A.Prompt template
2. B.Hallucination
3. C.System prompt
4. D.Reranking

Answer + AI explanation with Pro

1. A.System prompt — the leading instruction that sets an LLM's role, constraints and tone for a session
2. B.System prompt — an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce
3. C.System prompt — a store that indexes embeddings for fast approximate nearest-neighbor similarity search
4. D.System prompt — fetching the k most similar chunks to the query to place in the prompt context

Answer + AI explanation with Pro

1. A.an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce
2. B.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
3. C.fetching the k most similar chunks to the query to place in the prompt context
4. D.input/output filters that block unsafe, off-topic or policy-violating LLM content

Answer + AI explanation with Pro

1. A.Hallucination
2. B.Top-k retrieval
3. C.Embedding
4. D.RAG

Answer + AI explanation with Pro

1. A.Hallucination — a store that indexes embeddings for fast approximate nearest-neighbor similarity search
2. B.Hallucination — input/output filters that block unsafe, off-topic or policy-violating LLM content
3. C.Hallucination — Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
4. D.Hallucination — an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce

Answer + AI explanation with Pro

1. A.a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
2. B.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
3. C.an LLM producing fluent but factually wrong or unsupported output, which grounding aims to reduce
4. D.fetching the k most similar chunks to the query to place in the prompt context

Answer + AI explanation with Pro

1. A.Reranking
2. B.Vector database
3. C.Chunking
4. D.Guardrails

Answer + AI explanation with Pro

1. A.Reranking — a store that indexes embeddings for fast approximate nearest-neighbor similarity search
2. B.Reranking — a second-stage model that reorders retrieved candidates by relevance before they enter the prompt
3. C.Reranking — a dense numeric vector representing text so that semantic similarity becomes vector distance
4. D.Reranking — Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time

Answer + AI explanation with Pro

1. A.input/output filters that block unsafe, off-topic or policy-violating LLM content
2. B.Retrieval-Augmented Generation: grounding an LLM's answer in documents fetched from a knowledge base at query time
3. C.a store that indexes embeddings for fast approximate nearest-neighbor similarity search
4. D.a parameterized prompt with slots for context and user input, versioned like code

Answer + AI explanation with Pro

1. A.Reranking
2. B.Hallucination
3. C.Prompt template
4. D.Guardrails

Answer + AI explanation with Pro

1. A.Guardrails — the leading instruction that sets an LLM's role, constraints and tone for a session
2. B.Guardrails — a store that indexes embeddings for fast approximate nearest-neighbor similarity search
3. C.Guardrails — splitting source documents into passages sized for embedding and retrieval
4. D.Guardrails — input/output filters that block unsafe, off-topic or policy-violating LLM content

Answer + AI explanation with Pro