NVIDIANVIDIA-Certified Professional: Agentic AI2026-1010.1 Build intuitive UIs with user-in-the-loop interaction.
Simple Calculator - Human in the Loop
Completed by opening the official lab
Why it matters
A review control is only real when the answer changes what the workflow does next. This example puts a person’s yes or no in front of a decision to spend more of the agent’s execution budget.
What to practise
Run nat run with configs/config-hitl.yml. The custom retry_react_agent workflow hits a LangGraph recursion limit, and the reusable hitl_approval_function then asks, through the Toolkit’s user interaction manager, whether to proceed. Answer yes once and no once. Compare a retry with a higher max_iterations against a run that ends with a fallback result. Notice that the approval function comes from the POR to Jira example and is wired in through configuration, not coded into the agent.
How it fits
The two answers are the explicit transitions that 10.1.1 asks for: one resumes with a changed limit, the other stops. The run waits in the session for the reply, which is 10.1.2’s synchronous review; decide whether your own deadline could absorb that wait.