Agentic hospital supply control tower

Predict shortages before they hit the bedside.

MedPack AI forecasts next-24-hour hospital supply demand, checks whether inventory is truly usable, ranks warehouse packing priorities, and generates a command-center action plan with escalation guidance.

Open Railway App View Data Flow GitHub Repository
24h demand forecast horizon
Zero PHI operational data only
Local agents zero-token decision committee

The problem: hospital stock is not always usable stock.

A hospital may appear to have enough inventory, but supplies can be expired, recalled, reserved, delayed, misplaced, or clinically unsuitable. MedPack AI turns prediction into operational action.

1

Forecast demand

Predicts next-day usage from patient volume, acuity, department, procedure counts, and recent consumption patterns.

2

Validate usable stock

Filters out inventory that exists on paper but should not be approved because of expiry, recall, reservation, or safety rules.

3

Escalate safely

Creates an action plan with owner, priority code, transfer options, substitute choices, and escalation window.

Data flow architecture

This diagram is embedded as pure SVG, not Mermaid, so it renders reliably on GitHub Pages and static portfolio sites.

1. Capture Collect operational hospital and supply-chain signals.
2. Predict Estimate 24-hour usage and coverage gap.
3. Decide Rank risk, packing priority, substitutes, and escalation.
4. Remember Write the action and decision event to audit memory.

Application pipeline

MedPack AI is built as a practical operations loop, not just a prediction screen.

Input packet Department, item, stock, patient load, acuity, usage, supplier delay.
ML forecast XGBoost/fallback logic predicts next-24-hour demand.
Risk controls Usable-stock rules filter expiry, recall, reserve, and unsafe inventory.
Agent review Local agent committee explains demand, inventory, packing, and safety.
Action card Command center outputs owner, priority, escalation window, and audit trail.

Dashboard evidence

These screenshots are pulled from the repository docs folder and will render on GitHub Pages.

Command center output screenshot
Command Center Priority code, escalation window, action card, and operational decision.
What-if simulator screenshot
What-If Simulator Stress-tests ED surge, ICU spike, supplier delay, flu season, and disruption scenarios.

Technology stack

Designed to be portfolio-ready, explainable, and deployable.

Python Streamlit Flask API XGBoost / ML fallback RAG / FAISS Sentence Transformers JSON Memory Railway Deployment No PHI Local Agent Committee