Graded Quiz: Advanced Retrievers for RAG :Advanced RAG with Vector Databases and Retrievers (BM RAG and Agentic AI Professional Certificate) Answers 2025
Question 1
Primary function of a LangChain retriever & how it differs from a vector store:
❌ Retriever = vector store for structured data
✅ Retriever returns documents based on an unstructured query; vector store stores documents
❌ Same function
❌ Retriever stores documents
Explanation:
A vector store holds embeddings; a retriever queries the store and returns relevant documents.
Question 2
How does a vector store–based retriever operate?
❌ Keyword matching
❌ Metadata tag matching only
✅ Uses vector representations to find closest matches
❌ Uses SQL queries
Explanation:
It performs similarity search using embedding vectors.
Question 3
Primary function of a multi-query retriever:
❌ Converts unstructured → structured
❌ Prioritizes parent docs
❌ Handles multiple queries simultaneously
✅ Refines a single query into multiple variations for better retrieval
Explanation:
The retriever generates multiple reformulated queries to improve recall.
Question 4
Difference between self-query and parent-document retriever:
❌ Self-query = multiple queries
❌ Self-query focuses on parent docs
✅ Self-query modifies the query; parent doc retriever handles hierarchical doc relationships
❌ Self-query = structured, parent-doc = unstructured
Explanation:
Self-query extracts structured filters (like metadata). Parent-document retrievers return parent-level docs.
Question 5
Keyword-based ranking retriever in LlamaIndex:
✅ BM25 Retriever
❌ Vector Index Retriever
❌ Auto Merging Retriever
❌ Recursive Retriever
Explanation:
BM25 is a classic keyword-based ranking algorithm.
Question 6
Best index for semantic retrieval:
❌ DocumentSummaryIndex
❌ KeywordTableIndex
✅ VectorStoreIndex
❌ TF-IDF Index
Explanation:
Semantic search requires embeddings; the vector index is designed for that.
Question 7
Which fusion strategy assigns higher scores to docs near the top of any list?
✅ Reciprocal Rank Fusion
❌ Query Fusion
❌ Distribution-Based Fusion
❌ Relative Score Fusion
Explanation:
RRF boosts documents that rank highly in any query’s results.
🧾 Summary Table
| Q No. | Correct Answer | Key Concept |
|---|---|---|
| 1 | Retriever returns docs; vector store stores docs | Retriever vs vector store |
| 2 | Vector similarity search | Vector-store retriever |
| 3 | Refines query into multiple variants | Multi-query retriever |
| 4 | Self-query modifies query; parent retriever uses hierarchy | Retriever types |
| 5 | BM25 Retriever | Keyword-based retrieval |
| 6 | VectorStoreIndex | Semantic search |
| 7 | Reciprocal Rank Fusion | Fusion strategy |