An HR agent that answers policy questions from your own documents and runs leave requests through AI + human approvals — natively inside Microsoft Teams.
The agent reads intent and handles it end to end — no menus, no forms, no portal.
Policy questions answered strictly from approved documents via semantic RAG, with the source cited — never the open web.
Guided intake, date validation, and working-days computation — then AI + human (HITL) approval, written straight to Azure SQL.
One signed-in conversation routes to a grounded answer or a fully tracked leave request.
Every leave request is scored on duration and the clarity of its reason, then routed to exactly the right level of approval — so nothing risky is ever rubber-stamped.
Approvers are never hard-coded — the agent resolves the employee's manager (and their manager) from the Entra ID directory at request time, always following the live org chart.
Short request with a clear, justified reason — approved instantly, employee notified, status written to Azure SQL.
Triggered by mid-length requests, an unclear reason, or a reason that doesn't match the duration.
Long absences escalate one level up — reviewed by the manager's manager with full request context.
HR documents are embedded into vectors and retrieved by semantic similarity — so the agent finds the right policy even when the question is worded nothing like the document.
Same login, same security, same trusted environment — data never leaves Microsoft 365 & Azure, so it stays governed and compliant by default.
The agent, its semantic RAG knowledge source, and the leave flow — wired to Azure AI, Azure SQL, and Entra ID.
Entra ID sign-in with agent-level RBAC — authorized employees only, each seeing just their own leave.
App Insights tracks self-service, auto-approval, and turnaround; Power BI surfaces HR hours saved. Shipped Dev → UAT → Prod.