Secure LLM inference: enforce access control on confidential data

This post details the architecture and implementation of adapter-based access control for LLM inference using the llm-d framework and LoRA adapters. It elaborates on the training phase (partitioning data, creating LoRA adapters) and the inference phase (authentication, authorization via OPA, semantic matching for adapter selection, routing to inference pools). It also provides a concrete example of per-role inference routing for a software engineer.