Deployment and Scaling
NCP-AAI · 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?