Profound University

Question 1

You run a template agent and the output doesn't quite match your needs. What is the recommended approach?

If you have any doubts, please go back and rewatch Lesson 21: Running Your First Agent.

Find a different template that better matches your requirements.

Delete the agent and build a new one from scratch.

Submit feedback to have the template improved by the platform team.

Identify the specific node producing the issue, modify it, and run the agent again to compare.

Question 2

You're building an agent that needs to:

Pull your citation data from Profound;
Scrape a competitor's page;
Have an LLM analyze the comparison;
Send the results to Slack;

Which sequence of node categories does this flow require?

If you have any doubts, please go back and rewatch Lesson 21: Running Your First Agent.

Research, Research, LLM, Research.

LLM, LLM, LLM, Integration.

Data, Research, LLM, Integration.

Data, Data, Research, Logic.

Question 3

An agent compares this week's competitor pricing pages against stored versions and reports changes. After each run, it overwrites the stored versions with the current scrapes. Why is this update step critical?

If you have any doubts, please go back and rewatch Lesson 22: Building an Agent from Scratch.

Each run advances the baseline, so the next run compares against the latest state rather than the original - making the agent longitudinal.

Without updating, the knowledge base files would become corrupted over time.

The knowledge base has a storage limit that requires old files to be replaced.

Updating triggers automatic notifications to team members about new data.

Question 4

An agent's LLM node returns a long paragraph describing pricing changes across four competitors. Downstream nodes need to determine which specific competitor changed and whether to trigger a Slack alert. What should you add to the LLM node?

If you have any doubts, please go back and rewatch Lesson 22: Building an Agent from Scratch.

A longer, more detailed prompt asking the LLM to be more organized in its response.

Structured outputs with boolean fields for each competitor and a summary string, making the output programmable for downstream nodes.

A second LLM node to parse the paragraph into usable fields.

A logic node before the LLM to pre-filter the data by competitor.

Question 5

You're building an agent that pulls citation data from Profound's data nodes. The results seem skewed, your brand appears far more often than expected in the citations. What's the most likely configuration issue?

If you have any doubts, please go back and rewatch Lesson 23: Agent Strategy: Your AEO Operating System.

The date range is too narrow, creating a biased sample.

You forgot to filter by prompt type - sentiment prompts (which mention your brand by name) are inflating the citation data.

The agent is pulling data from too many platforms simultaneously.

Your knowledge base baseline is outdated and needs to be refreshed.

Building Agents

Building Agents