A police evidence AI zone evaluates hybrid RAG where vectors stay on-prem and LLM weights sit in cloud. What trade-off should the architecture evaluation emphasize?
Select an answer to reveal the explanation.
Short Explanation
Vectors at home, LLM in cloud: you keep sensitive embeddings closer, but every RAG hop pays latency and needs a secure path. Name that trade-off honestly. Hybrid is not free lunch.
Full Explanation
A police evidence AI zone evaluating hybrid RAG where vectors stay on-prem and LLM weights sit in cloud should emphasize that on-prem vectors plus cloud LLM weights trade privacy locality against cross-site latency and connectivity needs. Sensitive embeddings can remain closer to the evidence boundary while generation runs remotely, but every retrieval/generation hop pays RTT and depends on a secure, reliable path.
Claiming on-prem vectors plus cloud LLM weights trade chunk embedding dimensionality against transformer parameter count size needs misses the architecture-level issue—the evaluation must weigh privacy locality against cross-site latency and connectivity, not merely model sizing. Claiming on-prem vectors plus cloud LLM weights trade retrieval audit log detail against evidence retention duration needs is irrelevant here—the evaluation must focus on the privacy-locality versus cross-site latency and connectivity trade-off, not logging detail versus retention policy. Asserting that on-prem vectors plus cloud LLM weights trade index refresh cadence against model version update frequency needs dodges the core architecture trade-off the evaluation must capture: privacy locality versus cross-site latency and connectivity, not lifecycle update timing.
Exam caveat: Domain 2.5 hybrid designs succeed when trade-offs are explicit; privacy locality for vectors does not erase connectivity, auth, and latency budgets for evidence workloads.
Operational check: measure end-to-end hybrid RAG latency from the evidence zone, verify vectors never leave the on-prem store, and confirm the cloud LLM path is encrypted and policy-controlled under load.
Cloud LLMs reduce on-prem compute but do not remove the need for secure connectivity: sensitive city data sent to the cloud still must be encrypted in transit and at rest. Public LLM APIs are not appropriate for police records and body-camera archives.
Prefer to study offline? Printable PDF study book + test engine ($7.50, unofficial). These questions stay free here.