A Complete Persona Pattern Example

Question 1

What two new expert tasks are added to the invoice agent?

Answer 1

Categorizing expenditures into predefined categories and checking invoices against purchasing rules.

Question 2

Why use expert tools instead of hardcoding rules?

Answer 2

It keeps the architecture clean, adaptable, and avoids embedding complex rule logic in code.

Question 3

What does the categorize_expenditure tool take as input?

Answer 3

A one-sentence description of the expenditure.

Question 4

What does categorize_expenditure return?

Answer 4

A single category name chosen from a fixed list of 20 categories.

Question 5

Why is a fixed category list useful?

Answer 5

It constrains outputs so the agent returns consistent, valid classifications.

Question 6

What persona is used for expenditure categorization?

Answer 6

A senior financial analyst specializing in corporate spending categorization.

Question 7

What is the purpose of the purchasing rules expert?

Answer 7

To validate whether invoice data complies with company purchasing policies.

Question 8

Why load purchasing rules from disk instead of hardcoding?

Answer 8

Policies can be updated without changing code, and the human-readable rules become the logic.

Question 9

Where are purchasing rules stored in the example?

Answer 9

config/purchasing_rules.txt

Question 10

What happens if the purchasing rules file is missing?

Answer 10

The tool defaults to “assume all invoices are compliant” (or another fallback rule).

Question 11

What does check_purchasing_rules return in the basic version?

Answer 11

A compliance result (true/false) and a brief explanation of issues.

Question 12

Why is free-text compliance output sometimes insufficient?

Answer 12

Other systems need predictable structure for logging, workflows, and reporting.

Question 13

What does prompt_llm_for_json add to the compliance tool?

Answer 13

Enforces a structured JSON output that conforms to a schema.

Question 14

What fields are required in the compliance JSON schema?

Answer 14

compliant (boolean) and issues (string).

Question 15

Why is schema-enforced output important for downstream systems?

Answer 15

It makes results machine-readable and easy to store, trigger workflows, or aggregate.

Question 16

What example rule triggers noncompliance in the case study?

Answer 16

Purchases over $5,000 require pre-approval.

Question 17

In the example invoice, why is the purchase noncompliant?

Answer 17

Total exceeds $5,000 and lacks required pre-approval.

Question 18

What additional policy example is given for consulting services?

Answer 18

Consulting fees over $10,000 require an SOW (Statement of Work).

Question 19

What does the invoice agent’s updated workflow include after extraction?

Answer 19

Summarize expenditure, categorize it, validate against policies, store results, and report summary.

Question 20

What is the one-sentence summary used for in the workflow?

Answer 20

As the input to the categorization expert tool.

Question 21

What is the role of function calling in this updated agent?

Answer 21

It lets the agent invoke tools reliably using structured tool calls rather than parsed text.

Question 22

Why does this example demonstrate “personas as tools”?

Answer 22

Categorization and compliance expertise are encapsulated in separate tools instead of the core agent prompt.

Question 23

What is the advantage of separating compliance into a tool?

Answer 23

You can update rules, prompts, and schema without rewriting the agent’s core goals.

Question 24

What is the risk mentioned even with structured extraction?

Answer 24

The LLM may still hallucinate, so production systems may need additional safeguards.

Question 25

What is the key architectural theme of this case study?

Answer 25

Modular expertise: keep core orchestration clean and push domain reasoning into specialized expert tools.