Handling retrieval failures gracefully in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why is it important to handle retrieval failures gracefully in AI systems?

Imagine an AI assistant that fetches information from a database. What is the main reason to handle retrieval failures gracefully?

A To provide a smooth user experience even when data is missing or delayed

B To ensure the system crashes quickly so developers notice the problem

C To ignore errors and continue without informing the user

D To always return empty results without explanation

💡 Hint

Think about how users feel when an app suddenly stops working or shows confusing errors.

Handling retrieval failures gracefully means the AI system can still respond helpfully or inform the user clearly, avoiding frustration or confusion.

❓ Predict Output

intermediate

What is the output when retrieval fails and fallback is used?

Consider this Python code snippet simulating a retrieval with fallback:

def retrieve_data(key):
    data_store = {'a': 1, 'b': 2}
    try:
        return data_store[key]
    except KeyError:
        return 'default_value'

result = retrieve_data('c')
print(result)

What will be printed?

A 'default_value'

B 2

C KeyError

D 1

💡 Hint

What happens if the key is not found in the dictionary?

The code catches the KeyError and returns 'default_value' instead of crashing.

❓ Model Choice

advanced

Which model architecture is best suited to handle missing data during retrieval?

You want an AI model that can still make reasonable predictions even if some input data is missing or incomplete. Which model type is best?

A Standard linear regression without imputation

B Feedforward neural network without dropout or masking

C Recurrent neural network with masking for missing inputs

D Decision tree that requires complete data

💡 Hint

Think about models that can ignore or skip missing parts of the input.

Recurrent neural networks with masking can handle missing inputs by ignoring them during processing, making them robust to retrieval failures.

❓ Metrics

advanced

Which metric best reflects graceful handling of retrieval failures in AI predictions?

An AI system sometimes returns fallback predictions when retrieval fails. Which metric helps measure if these fallbacks keep predictions reliable?

A Accuracy on only complete data samples

B Mean squared error including fallback predictions

C Training loss before deployment

D Number of retrieval failures logged

💡 Hint

Consider a metric that measures prediction quality including fallback cases.

Mean squared error including fallback predictions shows how well the model performs overall, including when retrieval fails and fallback is used.

🔧 Debug

expert

Why does this retrieval fallback code cause a runtime error?

Review this Python code snippet:

def get_data(key, fallback=None):
    data = {'x': 10, 'y': 20}
    try:
        return data[key]
    except KeyError:
        return fallback.upper()

result = get_data('z', fallback=None)
print(result)

What error occurs and why?

A TypeError because data[key] returns an int but fallback is None

B No error, prints None

C KeyError because 'z' is not in data and fallback is ignored

D AttributeError because fallback is None and None has no 'upper' method

💡 Hint

What happens if fallback is None and you call a string method on it?

Calling .upper() on None causes an AttributeError because NoneType has no such method.