LangGraph E-commerce Assessment | LangGraph for E-commerce

1

An e-commerce company wants to build a customer service system that can handle order inquiries, returns, and product questions. What's the best LangGraph architecture?

One large agent that handles everything A router agent that classifies queries and routes to specialized sub-agents Separate disconnected agents for each task A simple if-else statement

2

You're building an order processing workflow. In what order should these steps execute?

Payment → Inventory Check → Validation → Fulfillment Fulfillment → Payment → Validation → Inventory Check Validation → Inventory Check → Payment → Fulfillment Validation → Payment → Inventory Check → Fulfillment

3

What happens when a LangGraph node returns only {'status': 'completed'} without other fields?

All other state fields are deleted An error is thrown Only the 'status' field is updated; other fields remain unchanged The graph restarts

4

A customer service agent detects that a customer is 'angry' based on sentiment analysis. What's the appropriate LangGraph pattern?

Ignore the sentiment and continue normally Delete the conversation Route to a human handoff node via conditional edge Respond with a generic message

5

You need conversation history to persist across browser sessions. What's required?

Store messages in the LLM Use a checkpointer (PostgresSaver, MemorySaver) with consistent thread_id Restart the graph each time Store in browser localStorage

6

In a product recommendation agent, why use a reducer like Annotated[list, operator.add] for the messages field?

To delete old messages To ensure new messages are appended to history rather than replacing it To encrypt messages To compress the messages

7

Your inventory agent needs to trigger automatic reorders. What's the safest approach for high-value orders?

Always auto-approve all reorders Never auto-order, always require manual approval Auto-approve within limits, use interrupt() for high-value orders requiring manager approval Send an email and hope someone sees it

8

What's the purpose of an idempotency key in payment processing?

To make payments faster To ensure retrying the same request doesn't charge the customer twice To encrypt payment data To track inventory

9

A LangGraph workflow is interrupted for human approval. How do you resume it?

Create a new graph and start over Call invoke() with Command(resume=value) using the same config/thread_id Delete the checkpoint and retry Wait for it to auto-resume

10

Your customer service agent sometimes takes 30+ seconds to respond due to slow LLM calls. What's the best mitigation?

Increase timeout to 60 seconds Use streaming to show tokens as they're generated, plus implement fallback models Remove the LLM entirely Add more servers

11

For a multi-agent e-commerce system, where should you implement rate limiting?

Only on the database Only on the frontend On LLM calls, external API calls (payment, shipping), and API endpoints Rate limiting isn't needed for AI systems

12

What information should be included when using interrupt() for human approval?

Just say 'approval needed' Full context: order details, risk factors, customer info, and available actions The entire conversation history Nothing - humans will figure it out

13

You're debugging a production issue where orders fail sporadically. What observability approach helps most?

Add more print statements Distributed tracing (LangSmith) showing the full execution path with timing Check the logs once a day Ask customers what happened

14

A VIP customer's order is flagged by fraud detection. Your workflow routes to 'fraud_review' node. The fraud team clears it. What should happen next?

Start the order from scratch Cancel the order Resume the workflow from where it paused, continuing to payment/fulfillment Send it back to fraud detection

15

You're building a recommendation agent that needs to know a customer's past purchases. What's the best state design?

Query the database in every node Include purchase_history as a list in the state, populated at the start Store purchases in a global variable Let the LLM remember purchases

Launch Offer