Module 7 Quiz — RAG Pipelines

Loading quiz...

---

Answer Key (Instructor Reference)

MCQ Answers

#	Topic	Answer
1	Why RAG	LLMs have no access to your data at runtime
2	LLM behavior	Confidently fabricates plausible-sounding answers
3	RAG definition	An architectural pattern
4	RAG capabilities	Provides evidence at runtime without changing the model
5	RAG vs fine-tuning	When you need answers grounded in specific, changing documents
6	RAG + fine-tuning	Fine-tuning teaches HOW; RAG provides WHAT
7	Core components	Retriever, Prompt Builder, Generator, Validator
8	Component separation	Each component can be tested and swapped independently
9	Retriever role	Find relevant chunks from the knowledge base
10	Prompt builder role	Structures retrieved chunks and question into a prompt
11	Data flow	Vector similarity search / retrieval
12	Normalized embeddings	To enable cosine similarity via inner product
13	Parameter k	The number of documents to retrieve
14	Trade-off k	More context vs. more noise
15	Evidence-first	Placing retrieved context before the question with explicit grounding instructions
16	RAG prompt instruction	If the context doesn’t contain the answer, say so
17	Chunk formatting	Format them clearly with labels (e.g., [1], [2])
18	Near-miss	A chunk that is semantically similar but factually different
19	Near-miss danger	They cause high confidence in wrong answers
20	Grounded hallucination	A wrong answer based on incorrect retrieved context
21	No relevant chunks	Refuse to answer
22	Score threshold	Rejects retrieved chunks below a similarity threshold
23	Enterprise refusal	In regulated environments, refusal is risk management
24	Context overflow	Too many chunks causing the model to lose focus
25	Precision@k	Proportion of retrieved chunks that are relevant
26	Faithfulness	Whether the answer only uses information from the provided context
27	Caching	Document embeddings, query embeddings, and optionally retrieval results
28	Latency dominant	LLM generation
29	Audit trail	Query, retrieved chunk IDs, scores, prompt, response, and timestamps
30	Auditability importance	To support debugging, compliance, and trust
31	Platform RAG	Supports self-service corpora, configurable policies, and full observability
32	RAG limitations	Guarantee correctness

Written Question Themes

#	Topic	Key Themes Expected
1	Why RAG needed	Runtime access to private/current data, hallucination prevention
2	RAG vs fine-tuning	RAG for data, fine-tuning for behavior
3	Architectural pattern	Components, testing, swappability
4	Data flow	Query → embed → retrieve → prompt → generate → validate
5	Near-miss	High similarity, wrong facts, dangerous
6	Failure modes	Near-miss, missing chunks, conflicts, overflow
7	Refusal as feature	Risk management, regulated industries
8	Prompt builder	Evidence-first, grounding, structure
9	Guardrails	Thresholds, validation, filtering
10	k trade-off	Context vs noise, filtering
11	Retrieval evaluation	Precision@k, relevance, MRR
12	Faithfulness	Context-only, no invention
13	Audit trails	Components, compliance, debugging
14	Caching strategy	What to cache, invalidation
15	Latency breakdown	LLM dominates, caching implications
16	Risk shift	Retrieval errors vs model hallucination
17	RAG platform	Multi-tenant, governance, observability
18	Regulated RAG	Extra guardrails, compliance, auditability