Generated from the build. Dev scope, synthetic SOP data.

Purpose

A staff-facing assistant in Microsoft Teams that answers policy and procedure questions from an organization's own SOPs rather than a model's general training. One router agent classifies each question and delegates to the right role-scoped specialist. Each specialist answers only from its own knowledge base and declines when its sources do not cover the question.

Components

• Concierge Router (n8n AI Agent): classifies the message and delegates to one specialist. Backed by a shared Claude model and conversation memory.
• Specialist agents (n8n AI Agent Tool): one per domain, attached to the router as tools. Example configuration: Frontline Operations, People and HR, Finance and Procurement, IT and Access.
• Per-domain knowledge bases (vector store, retrieve-as-tool): one store per domain, searched only when a question calls for it.
• Ingestion workflow: chunks and embeds each SOP corpus into its domain store.
• Teams delivery: an Azure Bot Service messaging endpoint posts to an n8n webhook, with an asynchronous proactive reply.

Flows

1. A staff member asks a question in Teams.
2. Azure Bot Service posts the activity to the n8n webhook, which returns 200 immediately.
3. The router classifies the question and calls the matching specialist.
4. The specialist retrieves from its knowledge base and drafts a grounded answer, or declines if undocumented.
5. The answer is posted back to Teams as a proactive message.

Dependencies

• n8n (AI Agent, AI Agent Tool, vector store, recursive text splitter).
• Anthropic Claude API (reasoning model).
• OpenAI API (embeddings).
• Microsoft Teams via Azure Bot Service and the Bot Framework Connector.