Agentic AI Healthcare Operations Capstone

SafeStaff AI

An emergency-department control tower that connects wait-time forecasting, operational pressure signals, nurse-staffing recommendations, human approval, and audit logging.

Created by Wil Low / Draculess99 as a final capstone project for the Google/Kaggle Agentic AI course.

XGBoost ER wait-time risk forecasting
Agentic AI Multi-agent staffing workflow
Human Approval Governance and audit trail

What SafeStaff AI Solves

Emergency departments face connected patient-flow failures: longer waits, rising operational pressure, nurse fatigue, boarding delays, and staffing gaps.

SafeStaff AI treats wait-time forecasting and nurse staffing as one connected workflow, helping decision-makers see when additional coverage may be needed.

Architecture & Data Flow

SafeStaff AI connects operational ER inputs, nurse registry data, shift schedules, policy/SOP grounding, wait-time forecasting, multi-agent review, human approval, roster action, and audit logging into one decision-support control tower.

SafeStaff AI architecture and data flow diagram showing inputs, Streamlit frontend, Flask backend decision engine, XGBoost forecast, RAG retrieval, multi-agent committee, human approval, roster update, audit log, and feedback loop

Architecture view: ER scenario inputs flow through the Streamlit control tower, backend forecast and agentic decision engine, then into human-governed staffing actions with audit-log feedback.

Key Features

๐Ÿ“ˆ

Wait-Time Forecasting

Uses XGBoost to estimate ER wait-time risk from operational features.

โš•๏ธ

Staffing Support

Translates wait-time risk and pressure signals into nurse recommendations.

๐Ÿง 

Agentic Workflow

Runs planner, compliance, safety, finance, and final arbiter reasoning.

โœ…

Human Approval

Keeps supervisors in control before roster changes are accepted.

๐Ÿงพ

Audit Logging

Records recommendations, approvals, rejections, overrides, and token mode.

๐Ÿ”’

Prototype Guardrails

Supports local deterministic mode and Live Gemini mode with fallback behavior.

How SafeStaff AI Works

The app takes emergency-department scenario inputs, forecasts wait-time risk, applies operational pressure signals, and routes staffing recommendations through a human approval workflow.

1

Scenario Inputs

Arrival surge, boarding, acuity, call-outs, fatigue, and shift context.

2

ML Forecast

XGBoost predicts ER wait-time risk.

3

Pressure Engine

Operational modules adjust the staffing recommendation.

4

AI Committee

Agents review staffing, safety, compliance, cost, and final decision.

5

Approval & Audit

Human approves, rejects, or overrides; the decision is logged.

Technology Stack

Python Streamlit Flask XGBoost Gemini API Railway Agentic AI Audit Logging

Why It Matters

SafeStaff AI demonstrates how machine learning and agentic AI can support healthcare operations by connecting wait-time risk, real-world ER pressure, and staffing decisions.

Prototype notice: This is a decision-support demo, not a clinically validated staffing system. Real deployment would require hospital governance, security review, validation, and human supervision.