LLM Assistant Pattern for Regulated Finance

Retrieval-grounded, validated, human-approved — never autonomous over money

LLM Assistant Pattern for Regulated Finance Retrieval-grounded, validated, human-approved — never autonomous over money governed inference zone Analyst · human in the loop · Architecture component Analyst human in the loop Guardrails · PII + schema + policy · governed inference zone · validation Guardrails PII + schema + policy validation LLM Orchestrator · prompt + tool routing · governed inference zone LLM Orchestrator prompt + tool routing Knowledge Base · vector + document index · governed inference zone Knowledge Base vector + document index Tools / Data Sources · read-only APIs · Architecture component Tools / Data Sources read-only APIs Audit Log · immutable evidence · governed inference zone · append-only Audit Log immutable evidence append-only prompt / document validated input retrieve context read-only APIs trace + citations policy decisions Legend Frontend Security Backend Database External

Grounding

  • • Answers cite retrieved internal documents
  • • The model summarizes context, it does not invent facts

Guardrails

  • • PII redaction on the way in
  • • Schema and policy checks on the way out
  • • Human approval before any consequential action

Evidence

  • • Every decision writes an append-only audit record
  • • Tool access is read-only and off the write path

Architecture diagram • Built with Archify • Create yours ↗ • Hover to trace • R route • Click to focus • +/− zoom • M radar • [/] views • P play story • T theme • E export