01 — COMMAND CENTER
Hospital flow visibility
Review simulated unit pressure, occupancy, pending discharges, delayed cases, and ED boarding from one operational view.
Hospital Flow Intelligence
Move patients safely. Recover beds faster.
BedFlow AI turns discharge-risk predictions into a prioritized, role-aware workflow—connecting operational blockers, assigned tasks, human accountability, capacity simulation, and auditable decisions in one command center.
XGBoost scoringRole-based workflowFHIR-shaped exportHuman decision control
Prioritized Discharge Queue
Model-scored operational review
Ready
Occupancy87%Simulated hospital
view
ED Boarders22Capacity pressure
Potential Beds8Scenario
estimate
PatientRiskDelayAction
PT0447Critical25.1hEscalate
PT0235Critical22.2hPlacement
PT0490High21.5hCase review
PT0166High19.8hHome care
Core capabilities
From prediction to accountable action.
BedFlow AI is not just a model dashboard. It converts predictions into prioritized work, assigns responsibility, monitors escalation, supports human decisions, and preserves an audit trail.
02 — PREDICTIVE ANALYTICS
Three-model scoring
Estimate discharge-delay probability, expected delay hours, and 30-day readmission risk using saved XGBoost models.
03 — WORKFLOW
Tasks and escalation
Turn incomplete readiness items into owned tasks with priorities, service levels, progress notes, and immutable event history.
04 — AGENTIC REVIEW
Safety versus flow debate
Compare patient-safety and operations perspectives before presenting a final recommended action plan.
05 — HUMAN CONTROL
Decision and accountability
Authenticated reviewers can accept, hold, escalate, override, or request more information with reasons captured for audit.
06 — INTEROPERABILITY
FHIR-shaped export
Package patient, encounter, observation, task, care-plan, and location information into a de-identified bundle.
Operational workflow
A complete discharge-flow decision path.
Every selected patient moves through the same transparent sequence—from model scoring and checklist review to task ownership, agent recommendation, and final human accountability.
STEP 01Select PatientChoose a case from the
prioritized operational queue.
STEP 02Score RiskRun the three saved XGBoost
prediction models.
STEP 03Review ReadinessInspect completed and
unresolved discharge requirements.
STEP 04Assign WorkCreate role-owned tasks with
urgency and escalation logic.
STEP 05Agent ReviewCompare safety, policy, and
hospital-flow perspectives.
STEP 06Human DecisionRecord the final
accountable action and rationale.
Decision intelligence
Understand the risk. Test the response.
The platform pairs patient-level model scoring with a capacity what-if simulator so operational leaders can explore where intervention may create the greatest benefit.
Predictive model stack
The same patient features are evaluated by three independently trained models.
Discharge Delay Risk
How likely is a discharge delay?
Classifier
How likely is a discharge delay?
Expected Delay Hours
How long could the delay be?
Regressor
How long could the delay be?
30-Day Readmission Risk
How likely is near-term readmission?
Classifier
How likely is near-term readmission?
Capacity What-If Simulator
Apply hypothetical operational changes, rescore the affected queue, and compare estimated impact.
Current State
Critical cases11
Delay hours168h
Open beds30
→
Scenario Estimate
Critical cases7
Delay hours104h
Potential beds38
Platform architecture
Built as a complete operational prototype.
The system combines a Streamlit experience, Flask APIs, trained model artifacts, role-aware workflows, persistent JSON storage, interoperability, and observability.
Technology stack
BedFlow AI keeps prediction, operational workflow, human decision-making, and integration clearly separated.
Streamlit ExperienceCommand center, workflow,
simulator, review
Flask API LayerAuth, predictions, tasks, audit,
FHIR, metrics
Decision IntelligenceXGBoost models + agentic review
+ guardrails
Operational StatePersistent JSON tasks, events,
audits, simulations
Deployment & ObservabilityDocker, Railway, request
IDs, health, readiness
Responsible-use boundary
BedFlow AI is a portfolio decision-support prototype using synthetic/proxy discharge-flow data and a public readmission dataset. It does not independently discharge patients, replace clinical judgment, or represent a validated production hospital system.
Explore the project
Healthcare operations intelligence, built beyond the model.
BedFlow AI demonstrates how predictive analytics can become an accountable operational workflow—not just another score on a dashboard.