Question 1

Example of Hub-and-Spoke pattern:

❌ Retriever → Summarizer → Compiler (this is a pipeline)
❌ Identical roles (redundancy, not hub-and-spoke)
❌ Independent agents (no coordination)
✅ A central Content Manager assigns tasks to Writer, Fact-Checker, SEO Optimizer

Explanation:
Hub-and-Spoke = one central coordinator (“hub”) delegating tasks to many specialized “spokes.”

---

Question 2

Summarizer Agent trying to fetch documents—root cause?

✅ Capability boundaries are not clearly defined, causing role confusion
❌ Missing generalist
❌ Compiler misintegration
❌ Wrong collaboration pattern

Explanation:
Agents must have clear scopes; otherwise they overstep roles (e.g., summarizer acting like retriever).

---

Question 3

Why is Interface Standardization important?

❌ Helps perform multiple roles
✅ Enables structured communication (e.g., JSON) between agents
❌ Removes need for orchestration
❌ Improves vector DB memory

Explanation:
Agents must pass info in predictable formats so the next agent knows how to process it.

---

Question 4

AI sent sensitive patient data—what safeguard was missing?

❌ Role-based access control
✅ PII sanitation at the model/orchestration layer
❌ Interruptability
❌ Loop detection

Explanation:
Sensitive data must be sanitized/filtered before an agent outputs or sends it externally.

---

Question 5

Why is autonomy a risk amplifier?

❌ Increases hardware needs
❌ Always leads to violations
✅ Reduces human oversight → increases misinformation, errors, security risks
❌ Eliminates repetitive errors

Explanation:
Autonomous agents make decisions unsupervised, amplifying risks when misaligned.

---

Question 6

Primary function of vector DB in RAG:

✅ Retrieve relevant information to provide context to the LLM
❌ Store past interactions
❌ Train the model
❌ Generate final answer

Explanation:
Vector DB stores embedded docs and performs similarity search for context.

---

Question 7

New role of LLM in agentic RAG:

❌ Filter queries
✅ Acts as an agent—deciding which data source or tool to call
❌ Encrypt data
❌ Create training data

Explanation:
In agentic RAG, the LLM chooses retrieval strategies, tools, and actions—not just generating text.

---

🧾 Summary Table