1. What should Ricky do before embedding documents in RAG?

❌ Select the chunking approach
❌ Set up the vector store
✅ Load and organize travel-related documents
❌ Connect retriever to the query engine

Explanation:
Before chunking or embedding, documents must first be loaded and organized.

---

2. What does context length refer to in LLMs?

❌ Long training time
❌ Amount of text to generate
✅ Large input lengths during training
❌ Vocabulary size

Explanation:
Context length = how many tokens the model can take in at once.

---

3. LangChain system crashes due to network interruption. What to do?

✅ Implement retry mechanisms
❌ Load all documents
❌ Increase chunk size
❌ Manually verify API

Explanation:
Retries handle temporary failures automatically and prevent crashing.

---

4. App skips important content between split sections. What to adjust?

❌ Chunk size
✅ Chunk overlap
❌ Length function
❌ Separator

Explanation:
Overlap preserves context between consecutive chunks.

---

5. Preserve document structure based on headers?

❌ RecursiveCharacterTextSplitter
❌ CSVRowSplitter
✅ MarkdownHeaderTextSplitter
❌ CodeTextSplitter

Explanation:
MarkdownHeaderTextSplitter splits text according to hierarchical header structure.

---

6. Relationship between retriever and vector store

❌ Vector store handles all text using LLM queries
✅ Retriever is an interface that returns chunks from vector stores
❌ Retriever generates vectors
❌ Vector store replaces retrievers

Explanation:
The vector store holds embeddings; the retriever queries it to fetch relevant chunks.

---

7. Sara wants complete document sections, not small chunks. Use:

❌ Self-Query Retriever
❌ Vector Store Retriever
✅ Parent Document Retriever
❌ Multi-Query Retriever

Explanation:
Parent Document Retriever returns the full parent section instead of small chunk slices.

---

8. Need detailed PDF metadata for each page → choose:

❌ MarkdownHeaderTextSplitter
❌ WebBaseLoader
✅ PyMuPDFLoader – provides rich metadata
❌ PyPDFLoader

Explanation:
PyMuPDFLoader returns page-level metadata like coordinates, fonts, blocks, etc.

---

9. Basic web interface demo for QA bot:

✅ Gradio
❌ TensorBoard
❌ React.js
❌ Flask

Explanation:
Gradio lets you quickly create interactive web UIs for ML demos.

---

10. Why does the QA bot answer accurately with citations?

❌ Knowledge graph
✅ Combines retrieval + LLM generation + embedded documents
❌ Prompt changes
❌ LLM memory

Explanation:
RAG = Retrieve → feed to LLM → generate answer grounded in sources.

---

🧾 Summary Table

---