Governance · Evaluation · Compliance
Overview
Enterprise Agentic AI cannot be completed with feature implementation alone. A governance framework is required that holistically manages quality evaluation (Ragas), operational playbooks, AI Gateway guardrails (PII and Prompt Injection defense), regulatory compliance (SOC2 and ISMS-P mapping), and domain customization strategies. This section addresses these five pillars in dedicated documents.
Document List
📄️ Ragas Evaluation
RAG pipeline quality evaluation and continuous improvement using Ragas
📄️ Agentic Playbook
Guide for declaratively defining agent workflows like IaC and automating compliance
📄️ AI Gateway Guardrails
LLM Gateway-level Guardrails — PII redaction, prompt injection defense, content filtering, tool comparison, and Korean financial compliance mapping
📄️ AI Gateway Multi-Tenancy
LLM Gateway-level multi-tenancy strategy — LiteLLM virtual key hierarchical model vs Kong Consumer policy comparison, budget enforcement, 3-tier tenant isolation (gateway, data, observability)
📄️ LLM FinOps Chargeback
LLM platform FinOps methodology — token metering, showback/chargeback strategies, agentic cost models, budget policies, and gateway integration
📄️ Compliance
Compliance guide mapping SOC2, ISO27001, Electronic Financial Supervisory Regulation (전자금융감독규정), and ISMS-P to AI operations
📄️ Domain Specialization
Guide to improving technical domain coding quality with LoRA Fine-tuning, VectorRAG, and GraphRAG — including FSI SI production scenarios