AI Staff Copilot
2026 · LLM staff concierge · Prototype / demo
A front-desk copilot that answers only from approved SOP manuals - and refuses when the corpus does not support an answer.
Operations UI spans the copilot, SOP browser, and a system monitor. FastAPI + LangGraph orchestrate the flow; SOP retrieval stays behind an MCP server over stdio.
Delivered
grounded answers with cited SOP source, refusal path, PII masking before orchestration, request receipts, SOP browser, and a system monitor.
Problem
Front-desk teams need fast answers under pressure, but generic assistants improvise policy and make operations riskier.
Design a hotel-ops copilot that only answers from approved SOP manuals, exposes exactly what policy context it used, and refuses when the corpus does not support an answer.
Intervention
Keep the UI thin in React, run orchestration in FastAPI + LangGraph, and isolate SOP retrieval behind an MCP server over stdio. Every request returns a receipt with request ID, timestamp, retrieved documents, and tracing mode.
Constraints
- No invented hotel rules when retrieval returns nothing relevant
- PII must be masked before the orchestration graph runs
- Operations staff need visible receipts and system health, not a black-box chat reply
Result
Operators can ask late-checkout, VIP, passport, and upgrade questions in one surface, verify the exact SOP source, and audit the full pipeline when support or compliance needs to inspect a decision.
A production copilot is safer when retrieval, refusal, masking, and receipts are product features, not invisible implementation details.
Screen breakdown
Start a conversation
Tell us who you are and what you need in a few lines. We reply with whether Iron Signal Works is the right fit and what a sensible first step looks like. Ranges and terms sit in pricing & terms.