NCP-AAI practice questions
NVIDIA · NCP-AAI · 200 questions
Original practice questions for the NVIDIA-Certified Professional: Agentic AI (NCP-AAI) exam, covering agent architecture and design, agent development, evaluation and tuning, deployment and scaling, cognition/planning/memory, knowledge integration and data handling, NVIDIA platform implementation (NIM, NeMo Guardrails, NeMo Agent Toolkit, TensorRT-LLM, Triton), run/monitor/maintain, safety/ethics/compliance, and human-AI interaction and oversight — framed in civic and municipal scenarios such as city AI pilots, permitting chatbots, 311 assistants, and public-health records.
This course contains the use of artificial intelligence.
About the NCP-AAI exam
- Time allowed
- 2 hours
- Questions
- 60-70
- Languages
- English
- Format
- Online, proctored remotely
Exam details published by the vendor, checked 6 November 2026. Vendors change fees and formats without notice — confirm on the vendor's own page before you book.
Practice Quizzes
Test your knowledge with standard 20-question practice sets.
Quiz 1
Quiz 2
Quiz 3
Quiz 4
Quiz 5
Quiz 6
Quiz 7
Quiz 8
Quiz 9
Quiz 10
Browse by Domain
Study specific topics at your own pace.
Agent Architecture and Design · 31 questions
- A city 311 portal redesigns its resident chat so callers can see what the agent is about to do before it files a service request. Which UI pattern best supports that goal?
- A permitting office wants counter staff to watch an agent draft checklist items while a homeowner sees only a simplified status strip. What architecture choice best protects that dual-audience design?
- A library chatbot adds quick-reply chips for hours, holds, and fine disputes instead of relying on free-text alone. When does that structured UI affordance most improve human–agent clarity?
- A transit authority mobile app shows the agent's next planned tool call as a readable step before booking a paratransit ride. What design principle does that preview primarily support?
- A county fraud-review agent must alternate between reasoning about a benefits claim and calling eligibility APIs. Which approach best fits that tool-using workflow?
- A water-utility outage bot keeps looping Thought→Action→Observation until it can cite a work-order ID. Which ReAct step is primarily responsible for grounding the agent's advice in the tool result?
- Emergency-management planners compare a ReAct-style agent to a static FAQ decision tree for storm-debris pickup guidance. When is the reason–act framework the better architectural choice?
- A clerk's agenda agent proposes actions only after writing an intermediate rationale the auditor can later read. Why make those reasoning traces an architectural requirement?
- A permitting planner agent must hand a parcel ID and zoning flags to a fee-calculation agent without free-form chat noise. What communication design best supports that handoff?
- Public-health case agents share only de-identified encounter codes when asking a scheduling agent for clinic slots. What does that protocol choice primarily enforce?
- Two 311 triage agents disagree on priority and the architecture needs a defined way to escalate. Which collaboration pattern element is most important to add?
- A transit disruptions agent must notify both a rider-notification agent and an ops-dashboard agent when delays occur. Which communication style best fits that citywide event fan-out?
- A multi-turn housing-intake chat must remember the current session's income answers without permanently storing them. Which memory design matches that requirement?
- A parks reservation agent stores recurring group preferences across seasons but expires one-off picnic details. What memory policy best describes that architecture?
- Court-clerk assistants lose case context when the token window rolls, so architects add episodic memory retrieval. What problem does that architectural layer primarily solve?
- A benefits agent wrongly recalls last year's household size because long-term memory lacked version stamps. What memory design change best prevents that class of error?
- A city AI pilot coordinates intake, eligibility, and notification agents for rental-assistance applications. Which orchestration pattern best fits an accountable end-to-end civic workflow?
- Snow-route planning uses a planner agent plus validator and dispatcher agents that share a task board. What does the shared workflow state primarily enable?
- Permit review stalls when two specialist agents wait on each other indefinitely. What coordination design most directly prevents that deadlock?
- A school-district IEP assistant routes speech, OT, and admin agents with clear ownership of each step. Why is role specialization important in that orchestration?
- Business-license renewal needs a chained sequence: validate entity → check taxes → generate invoice draft. Which architectural approach best implements that ordered municipal process?
- Code-enforcement triage uses a logic tree for severity then a prompt chain for remedy letters. What does combining those techniques provide?
- A grant-writing helper must keep workflow state across days so staff can resume mid-draft. Which orchestration capability is essential?
- Assessor appeal prep must branch if comparable sales are missing versus if comps exist. How should stateful orchestration handle that?
- Zoning agents must reason over parcel–use–ordinance relationships. When is a knowledge graph a better architectural fit than flat document RAG alone?
- A public-works outage agent must answer which hydrants lose pressure when a water main is isolated, using links among mains, hydrants, and open work orders. Why should the architecture integrate a knowledge graph rather than flat document retrieval alone?
- A social-services agent must spot when a household is enrolled in two programs that conflict under city policy. What knowledge-graph design choice best supports that relational conflict check?
- City attorneys want a research agent to follow ordinance citation links for precedent lookup without inventing case law. How should knowledge-graph integration support that relational reasoning?
- Summer tourism spikes 311 volume. Architects need more agent capacity without redesigning every tool integration. Which adaptability and scalability approach best fits?
- Public works launches a new stormwater fee program and needs a specialist agent in the existing multi-agent stack. What architectural adaptability pattern should the team use?
- County IT fears a single monolithic tax-season agent will cascade failures across filing, payment, and FAQ tasks. Which architecture best improves failure isolation and scalability?
Agent Development · 31 questions
- A permitting bot must cite ordinance section IDs before advising residents on setbacks. Which prompt-engineering practice most improves that reliability?
- A 311 routing agent must consider today’s street closures before choosing a crew queue. How should dynamic prompt chaining support that?
- Housing staff notice empathy and next-step structure drift across languages in the assistance chatbot. What prompt-engineering fix best improves reliability?
- A clerk agent’s prompt chain fails because step 2 expects fields that step 1 never produced. What development practice prevents that failure mode?
- A library FAQ agent over-answers legal and medical questions outside published services. Which prompt practice keeps it inside service boundaries?
- Residents submit pothole photos that must become structured work orders. How should the city integrate generative and multimodal models in the agent workflow?
- A 311 phone line converts speech to text before an LLM agent triages the call. Where should audio models fit in that multimodal pipeline?
- Building inspectors upload annotated photos, and the agent must not invent defects that are not visible. Which multimodal reliability practice should developers enforce?
- Parks marketing wants one agent stack to draft event captions and generate flyer images. When is integrating generative multimodal capability the right development choice?
- Transit wayfinding agents must accept typed addresses and selfie-at-stop photos. How should developers design multimodal input handling?
- A benefits agent must query eligibility with validated case IDs. What custom-tool design best supports safe municipal system interaction?
- Permit fee amounts must stay accurate for residents. Which tooling pattern should the permitting agent follow?
- Emergency operations wants an agent tool that pages on-call staff through the city’s paging gateway. What is the critical development requirement for that tool?
- Resident-facing agents need GIS layers from the open-data portal but must not alter authoritative maps. How should tool permissions be designed?
- A clerk agent must use three legacy SOAP services for license renewal. What integration approach best fits custom tool development?
- During a storm, GIS lookup times out for a flooding advisory agent. Which error-handling approach is most appropriate?
- The permitting agent’s payment tool receives HTTP 429 responses from the cashiering service. How should error handling respond?
- Eligibility-service failures keep repeating, so residents need a clear fallback instead of endless retries. Which pattern should the benefits agent implement?
- A permitting workflow finds the parcel successfully but the fee service fails. How should the agent handle that partial multi-tool success?
- A GIS address lookup fails with a DNS timeout during a brief network blip, and the agent replies that the street address does not exist. What error-handling approach should developers apply?
- Residents watching a streamed 311 answer can tap “wrong street” while tokens are still arriving. How should the conversation flow be developed?
- Counter staff need to watch tokens stream while an agent drafts a denial letter so they can co-edit live. What development choice best supports that workflow?
- A housing chatbot should ask clarifying follow-ups when retrieval confidence is low. How should its dynamic conversation flow behave?
- Transit chat freezes until a complete answer is ready, and product owners want first-token latency SLOs. Why develop streaming conversation flows?
- An emergency FAQ agent is mid-stream answering a routine question when a higher-priority alert banner must take over. How should streaming be developed?
- After wrong triage of mold complaints, the city reviews which decision heuristic the agent used. What should developers do next?
- Two business-license routing strategies are candidates for production. How should the city refine which decision-making approach to keep?
- Inspectors report the agent over-prioritizes speed versus accuracy on safety tickets. How should decision strategies be refined?
- A grants agent preferred general web search over the city’s official policy API for award rules. How should its tool-selection strategy be refined?
- School-district agents escalate routine questions to humans too early, flooding counselors. What decision-strategy change is appropriate?
- A post-incident review finds the city agent gave legal advice because it lacked a refuse/escalate branch for out-of-scope questions. What decision strategy should be added?
Evaluation and Tuning · 27 questions
- City IT builds a gold set of 200 labeled 311 dialogues to score the agent weekly. What evaluation practice does this implement?
- Permit-answer accuracy must be measured against ordinance excerpts. Why is that benchmark choice better than generic chatbot leaderboards?
- Public-health agents must be scored on correct triage category and on refusing diagnosis. How should the evaluation pipeline be designed?
- A city RAG agent’s evaluation should distinguish bad retrieval from bad generation. What pipeline design achieves that?
- Nightly CI must run a frozen civic scenario suite before any agent prompt change merges. What evaluation practice is this?
- The same agent scores high on trash-pickup FAQs but fails on complex variance petitions. What evaluation insight should teams take?
- English and Spanish housing datasets show unequal agent success. How should performance be compared?
- Two retrieval corpora—municipal code versus website FAQ—are tested on the same permit questions. What is the evaluation goal?
- Seasonal storm datasets break a summer-tuned 311 agent. What does comparing performance across temporal slices reveal?
- Multi-agent and single-agent setups are candidates for rental-assistance completion. How should teams compare them fairly?
- Counter-staff assisted mode outperforms fully autonomous mode on appeal letters. What comparison should drive the operating choice?
- Residents can rate answers thumbs-up/down and must pick a reason code such as wrong hours, wrong fee, or unclear. Why collect feedback that way?
- Clerks file structured “agent erred” tickets that feed a weekly prompt-fix backlog. What evaluation-and-tuning practice is this?
- A transit agency clusters free-text rider complaints into themes for agent improvement. What should the team do first with that unstructured feedback?
- Feedback for a benefits-eligibility agent sometimes includes case numbers. What must happen before that feedback enters the training or eval store?
- A housing portal denies an application and explains why. Why require structured feedback right after that denial explanation?
- 311 leadership wants sub-two-second answers, but lowering max tokens truncates complete fee explanations. What trade-off should the tuning decision reflect?
- A city lowers temperature for ordinance quoting but keeps it higher for parks event brainstorming. What does that parameter choice recognize?
- GPU costs spike while the city runs off-peak permit batch jobs. How should tuning address the latency–cost–accuracy trade-off?
- A clerk agent sometimes invents unused ordinance IDs. Which tuning move best reduces that hallucination pattern?
- A larger model improves appeals accuracy but misses the city’s p95 latency SLO. What should the team do?
- Evaluation shows retrieval misses concentrated on alley easement documents. What is the best next optimization step?
- Confusion matrices show trash versus recycling tickets often misrouted. What should the team optimize first?
- Latency budgets fail only on multimodal photo-ticket turns. How should optimization proceed?
- Faithfulness scores drop after a new website scrape. Analysis points to noisy HTML boilerplate. What should the team do?
- Human review finds many answers that are factually correct but unhelpful. What optimization best improves civic usefulness?
- Week-over-week dashboards show Spanish-language task scores lagging English. What should longitudinal eval analysis drive?
Deployment and Scaling · 27 questions
- A city moves its multi-agent permitting stack from a laptop demo to a shared production cluster. What concern becomes newly critical at production scale?
- How should orchestrator, tools, and specialist agents be deployed for a production civic multi-agent system?
- Before full cutover, operations wants a safe rollout of a new 311 multi-agent graph. Which approach fits?
- Peak snow-day load requires autoscaling the planner agent. What must production orchestration preserve?
- A county disaster-recovery plan requires the multi-agent stack to survive an availability-zone outage. What deployment property is essential?
- A production gate blocks go-live until tool allowlists and secrets are handled correctly. Where should those controls live?
- For the clerk agent, how should prompt and tool-schema changes move to production?
- Before any production merge for the benefits bot, what governance control should MLOps enforce?
- What CI/CD practice best validates a new agent build before promotion using real municipal traffic patterns?
- Each agent release that affects residents must clear municipal change control. How should MLOps reflect that?
- Automated CI should fail an agent pipeline when which tool-governance condition is detected?
- A city’s agent deploy checklist is incomplete without which MLOps artifact?
- Before a transit agency launches its paratransit booking agent cluster, what load-profiling step best proves reliability under distributed demand?
- Tail latency spikes when every municipal agent calls the same GIS service. What should profiling focus on first?
- How should a city profile agent reliability before trusting a housing chatbot in production?
- When profiling a GPU-backed 311 agent fleet, why separate GPU utilization and queue depth from application RPS?
- Multi-region routing adds about 80ms to a city’s 311 chat path. What should latency profiling determine?
- How should specialist agents be packaged for scalable deployment on the city’s Kubernetes platform?
- Resident 311 sessions carry workflow state. What load-balancing approach fits containerized agent replicas?
- Inference-heavy municipal agent services need Kubernetes HPA. Which signals should drive scaling?
- A permitting office wants safe Kubernetes cutover for its agent with instant rollback. Which pattern fits?
- In the city’s zero-trust Kubernetes cluster, how should multi-agent pod traffic be secured at scale?
- City finance wants cheaper GPU nodes at night while emergency FAQ agents stay highly available. What approach fits?
- Where is it appropriate to use spot or preemptible nodes in a municipal agent platform?
- Evening peaks spike inference cost for repeated trash-schedule questions. What cost-aware tactic still preserves availability?
- Multi-AZ agent deployment roughly doubles cost. How should architects justify it for a public-facing city AI service?
- One department’s agent traffic is starving the shared city cluster. What control preserves fairness and availability?
Cognition, Planning, and Memory · 20 questions
- A multi-day variance hearing assistant must recall exhibits across sessions. What memory approach fits?
- A 311 agent must track the live ticket fields and also recall similar past tickets. How should those memory roles differ?
- Before a housing-intake agent writes to long-term memory, what privacy step should run?
- Token overflow drops a resident’s address mid-311 chat. What short-term memory fix should the team add?
- An appeals agent must handle “challenge this assessment.” Which reasoning approach fits best?
- Fee-logic agents need chain-of-thought internally. How should that reasoning appear to residents?
- A storm-response agent must restore service. Which reasoning pattern structures the work?
- Verbose chain-of-thought from a city attorney’s assistant begins showing draft legal theories to residents. What should change?
- A business-license agent must not skip tax clearance. What planning strategy enforces that gate?
- A building-permit agent planned an inspection for next Tuesday, but the scheduling tool reports no open slots that week. What should the multi-step planner do?
- A community-grant agent must collect bank letters, budgets, and board resolutions, then wait for a compliance officer’s sequential review. How should planning combine those phases?
- A demolition-request planner proposes scheduling a wrecking crew before checking historic-status flags in the parcel system. What planning defect does that expose?
- A rental-assistance workflow spans income verification, landlord confirmation, and payment authorization across agencies. Why should orchestration checkpoint after each verification step?
- During a zoning-questions session, the agent already retrieved three ordinance sections. How should shared orchestration state improve efficiency?
- A human clerk takes over a stalled business-license case mid-flow. What must stateful orchestration support when the agent resumes?
- A fee-payment step in a permit orchestration may retry after a timeout. What safeguard prevents charging the resident twice?
- A 311 trash-day agent repeatedly fails on holiday weeks. Based on that feedback, how should it adapt its reasoning strategy?
- Code inspectors repeatedly correct an agent that cites violations from undated photos. What experience-based adaptation should the agent adopt?
- Resident satisfaction scores drop on housing-assistance denial letters that feel abrupt. How should outcome feedback reshape the agent’s planning style?
- Permit reviews in certain ZIP codes often fail late when floodplain overlays are checked last. What adaptation should the agent learn from that geographic pattern?
Knowledge Integration and Data Handling · 20 questions
- A permit Q&A assistant must ground answers in the municipal code corpus. Which retrieval design best fits authoritative civic documents that need both topical and exact-match strength?
- Semantic-only search keeps missing queries that cite exact ordinance numbers like “12.04.210.” What change should the RAG pipeline make?
- During a flash-flood emergency, the city posts updated shelter FAQs that must reach the resident agent within minutes. What retrieval-pipeline requirement matters most?
- Clerks must audit which municipal-code section grounded a permit agent’s answer. What RAG design choice supports that audit?
- A city vector index holds policies spanning decades. How should the vector database be configured for precise permit retrieval?
- Long zoning chapters retrieve as fragmented, incoherent snippets. What vector-index optimization should the team tune first?
- p95 vector retrieval latency for the 311 agent breaches the published SLO. What is an appropriate vector-database response?
- Public website copy and confidential case notes must both support agent retrieval. How should the vector database be organized?
- A municipal agent needs current permitting, tax, and GIS data for reasoning. Which approach best integrates those heterogeneous sources?
- Assessor and utility systems use different parcel identifier formats for the same property. What ETL concern must be resolved before agent use?
- Housing-intake extracts contain Social Security numbers destined for an agent-facing store. What privacy-preserving ETL step should run first?
- The nightly ETL that refreshes fee tables fails, but a release pipeline still wants to promote a fee-quoting agent. What should happen?
- Preprocessing finds two ordinance sections with the same number but conflicting effective dates. What data-quality action should occur before embedding?
- Scanned city-council minutes enter the RAG corpus with OCR garbling of names and motion text. What should preprocessing do before indexing?
- Flood-complaint tickets are rare in the city’s training and eval sets, so the agent underperforms on storm scenarios. What judicious preprocessing step helps without inventing policy?
- A municipal agent misses parcel matches because residents write “St” while the assessor store uses “Street.” What preprocessing fix best improves civic entity matching?
- During a water-main outage, the city’s agent must reason over live work-order SQL status plus unstructured email notes from field crews. Which design best enables that?
- Transit ops wants live bus locations for the resident agent while service-policy PDFs remain in the vector store. How should knowledge access be architected?
- A public-health agent gives wrong clinic hours because it reads a stale cache. What should the city mandate for hours data?
- A court clerk agent must answer both “when is the hearing?” and “what did the opinion say?” Which knowledge pattern fits?
NVIDIA Platform Implementation · 14 questions
- The city wraps its resident agent so it cannot invent medical diagnoses. Which NVIDIA integration best enforces that compliance and safety control?
- Campaign-season questions to the city agent must not become political advice. How should NeMo Guardrails dialog policy handle them?
- A benefits agent must not send Social Security numbers into downstream tools. Which NeMo Guardrails control fits best?
- County IT needs high-performance LLM inference behind the 311 agent orchestrator. Which deployment choice best matches NVIDIA platform practice?
- A permit RAG agent needs both a chat LLM and an embedding model in production. How should NIM be used?
- Before resident traffic reaches a new NIM deployment, platform ops must know the microservice is healthy. What should be wired?
- Developers need to prototype a multi-agent housing-intake graph for the housing authority. Which NVIDIA surface best fits workflow optimization and development?
- NeMo Agent Toolkit profiling shows tool latency dominates a city’s permitting workflow. What should architects do next?
- The city wants an evaluation harness for its agents wired into NVIDIA agent toolkit surfaces. Which approach aligns?
- Live permitting chat misses its p95 latency SLO. Which NVIDIA optimization path is most appropriate?
- Evening 311 peaks need higher throughput across multiple agent models. Which Triton feature helps?
- Under the city’s latency SLO, unoptimized Hugging Face serving fails for the public agent. What should the team select?
- Field crews upload pothole photos that a multimodal agent must reason over. What NVIDIA-platform practice applies first?
- Municipal contact-center audio from 311 calls must become text before agents continue. How should the multimodal pipeline be optimized on NVIDIA platforms?
Run, Monitor, and Maintain · 10 questions
- Ops must watch the live 311 agent’s health beyond “is the pod up?” Which dashboard definition best fits?
- A civic AI dashboard already shows uptime. What additional reliability metrics should ops track for the agent itself?
- “Parcel not found” errors spike after a GIS change. How should ops find the root cause?
- A permit answer fails in production across orchestrator, NIM, and vector DB. What diagnostic approach helps most?
- Before a new 311 agent build takes full traffic, ops wants proof it is not worse than last week. What continuous benchmarking practice fits?
- English scores on the city’s 311 agent rose, but Spanish task scores dropped. What continuous benchmarking gate should block promotion?
- City MLOps stores prompt revisions for the permit agent. When should a new prompt version auto-promote to production?
- An embedding refresh makes ordinance retrieval worse for the municipal agent. What production practice limits damage?
- 311 tools start failing but the city still wants resident trust. What uptime and transparency practice fits?
- The live municipal agent invents a fake fee rule and residents notice. What ops practice maintains trust during the incident?
Safety, Ethics, and Compliance · 10 questions
- A municipal case-management agent calls tools that open resident case files. What security and audit design is required?
- The permitting agent can propose fee waivers. How should system security control who approves them?
- A public-health agent is asked to email full medical notes to a personal address. What compliance guardrail should apply?
- City records policy limits chat transcript retention. How should the 311 agent stack comply?
- Housing-assistance answers may differ unfairly by neighborhood. What bias-mitigation practice should the city apply?
- The public portal agent might generate abusive text. What toxicity control belongs in the path?
- A resident asks the building agent for structural engineering advice on a cracked beam. What layered safety approach fits?
- An emergency municipal agent detects life-safety keywords in a resident chat. What escalation protocol should run?
- The city wants to redistribute fine-tunes of an open-weight model used by its agents. What compliance check comes first?
- Procurement asks how the civic agent stack maps to emerging AI regulatory documentation duties. What response shows compliance awareness?
Human-AI Interaction and Oversight · 10 questions
- The permit portal agent drafts an application for a resident. What user-in-the-loop UI pattern should apply before submit?
- Clerks oversee the ordinance agent’s replies. Which HITL UI best supports their review?
- The 311 site shows agent suggestion chips. What inclusive UI check should pass before launch?
- The city wants resident feedback to improve the 311 agent over time. What structured loop fits?
- Field inspectors review agent-drafted violation notices. How should their feedback improve the agent?
- A resident questions a fee the agent quoted. What transparency mechanism should the UI offer?
- Auditors review a denial letter produced with agent help. What transparency artifact do they need?
- Residents want to know why the civic agent answered as it did. How should explainability balance openness and safety?
- A live 311 agent session goes off track. What human oversight capability should supervisors have?
- The finance agent recommends high-dollar resident refunds. What oversight gate protects accountability?
These questions are original practice material and are NOT actual exam questions or brain-dump content. All vendor marks are trademarks of their respective owners. This site is not affiliated with, endorsed by, or sponsored by the exam vendor.