An edge-deployed model fine-tuned on the company's domain: Which combined approach best fits all constraints?
Edge deployment keeps data on-site with low latency despite poor connectivity, and fine-tuning embeds the required specialized terminology, satisfying every constraint.
The question
A manufacturer needs an AI assistant at remote factories with unreliable connectivity and a hard rule that no operational data leaves the site; it must respond within milliseconds and consistently match the company's specialized terminology. Which combined approach best fits all constraints?
Preparing for AIGP? Take the free 5-min readiness quiz →
- A cloud-hosted model using retrieval augmented generation over site dataRAG can supply terminology, but cloud hosting sends data offsite and depends on the connectivity the scenario rules out.
- A cloud API model wrapped in an agentic multi-step orchestration layerAgentic orchestration adds capability, yet a cloud API still breaks the data-residency and offline-latency constraints here.
- An on-premise central server serving all factories over the shared WANIt keeps data in-house, but routing remote sites over an unreliable WAN fails the low-latency and connectivity requirements.
- An edge-deployed model fine-tuned on the company's domain terminology ✓Edge hosting keeps data on-site under poor connectivity with low latency, and fine-tuning bakes in the specialized terminology it must match.
The trap
Selecting a cloud or central-server option for capability while overlooking hard data-residency and low-latency-at-the-edge constraints. How to remember it
Edge deployment keeps data on-site with low latency despite poor connectivity, and fine-tuning embeds the required specialized terminology, satisfying every constraint.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Deployment and Use questions
- An open-source model self-hosted with full access: Which model-type choice best satisfies all three →
- To evaluate the potential harms the system could pose: What is the primary purpose of this activity? →
- Identifying and evaluating key terms and risks: Which assessment activity is this? →
- All 424 Understanding How to Govern AI Deployment and Use questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Deployment and Use ·
Every answer, right and wrong, comes with its own explanation.