Combining retrieval with agent reasoning in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does retrieval enhance agent reasoning?

Imagine an AI agent that answers questions by searching a large database of documents. How does adding a retrieval step improve the agent's reasoning process?

A It provides relevant information from external sources, allowing the agent to reason with up-to-date and specific data.

B It replaces the agent's reasoning entirely, so the agent only returns retrieved documents without processing them.

C It slows down the agent because retrieval adds unnecessary steps that do not affect reasoning.

D It limits the agent to only use pre-stored knowledge, preventing it from adapting to new queries.

💡 Hint

Think about how having access to more information can help an AI make better decisions.

Retrieval allows the agent to access relevant external information, which it can then use to reason and generate more accurate and context-aware answers.

❓ Predict Output

intermediate

Output of agent reasoning with retrieval

Given the following simplified agent code that retrieves documents and then answers a question, what is the output?

Agentic AI

documents = {"doc1": "The sky is blue.", "doc2": "Grass is green."}

query = "What color is the sky?"

retrieved = [doc for doc in documents.values() if "sky" in doc]

answer = "".join(retrieved) if retrieved else "No info found."

print(answer)

A Grass is green.

B The sky is blue.

C No info found.

D The sky is blue.Grass is green.

💡 Hint

Look for the document containing the word 'sky'.

The code filters documents containing 'sky' and joins them. Only 'The sky is blue.' matches, so that is printed.

❓ Hyperparameter

advanced

Choosing retrieval size for agent reasoning

When combining retrieval with agent reasoning, what is the effect of increasing the number of retrieved documents (retrieval size) on the agent's performance?

A Increasing retrieval size can improve accuracy but may increase processing time and introduce irrelevant information.

B Increasing retrieval size always improves accuracy without any downsides.

C Increasing retrieval size decreases accuracy because more documents confuse the agent.

D Retrieval size does not affect agent performance at all.

💡 Hint

Consider the trade-off between more information and processing cost.

Retrieving more documents can help by providing more relevant data but may slow down the agent and add noise, so a balance is needed.

❓ Metrics

advanced

Evaluating combined retrieval and reasoning agents

Which metric best measures how well an agent combines retrieval with reasoning to answer questions accurately?

A Precision of retrieved documents only.

B Recall of retrieved documents only.

C Number of documents retrieved per query.

D End-to-end answer accuracy comparing agent answers to ground truth.

💡 Hint

Think about what really matters: the final answer quality.

While retrieval metrics matter, the best measure is how accurate the agent's final answers are after reasoning with retrieved data.

🔧 Debug

expert

Debugging agent reasoning with retrieval failure

An agent retrieves documents but always returns 'No info found.' even when relevant documents exist. What is the most likely cause?

Agentic AI

documents = {"doc1": "Cats are Mammals.", "doc2": "Dogs are friendly."}

query = "What animals are mammals?"

retrieved = [doc for doc in documents.values() if "mammal" in doc]

answer = "".join(retrieved) if retrieved else "No info found."

print(answer)

A The join method is incorrect and causes an error.

B The documents dictionary is empty, so no retrieval is possible.

C The retrieval condition uses 'mammal' but documents contain 'mammals' (plural), causing no matches.

D The query variable is not used in retrieval, so retrieval always fails.

💡 Hint

Check if the retrieval condition matches the exact words in documents.

The retrieval looks for 'mammal' but documents have 'mammals' with an 's', so no documents match and the agent returns 'No info found.'