08:12
67
84
6
83

Concept & Idea Catalog

What Pulse is and why, who it's for, the full 27-module catalog, benefits by audience, and the responsible-AI gaps.

All docs

Pulse — Product Concept & Idea Catalog

This document is the thinking behind Pulse: what it is, the operating principles, the full module catalog (27 modules across 12 domains), a deep-dive on the four flagship modules built in this repo, and — importantly — a large set of additional ideas and the responsible-AI gaps that a real build must own. It was synthesized from a structured multi-agent exploration of hospital operations; the raw synthesis is in ideation/full-synthesis.json.

1. The thesis

A hospital is a real-time logistics system pretending to be a set of departments. The work of coordinating it — who goes in which bed, who covers the sick call, what's blocking this discharge, which waiting patient is quietly deteriorating — is done today by phones, pagers, whiteboards, spreadsheets, and a morning huddle that's stale by mid-morning. That coordination overhead is enormous, invisible, and exactly what software should absorb.

Pulse's job: ingest the live signals that already exist, predict where bottlenecks will form hours ahead, and close the loop — propose the specific action, route it to the right person with an SLA, and on accept, fan out the downstream work automatically. The measure of success is not dashboards; it's administrative overhead removed and throughput recovered.

Operating principles

  1. Orchestrate, don't just display. Every screen ends in an action, not a number to stare at.
  2. The brain is always visible and always overridable. Every recommendation ships its reasoning (factor × weight = contribution). Nothing auto-executes silently.
  3. One urgency language. A single ok · watch · urgent · critical scale means a dirty-too-long bed, an SLA-at-risk wait, and a ratio breach all read identically across every board.
  4. Close the loop and prove it. Track every recommendation proposed → accepted/overridden → executed → outcome, and tally the boarding-hours and agency shifts avoided. Trust is earned.
  5. Integration-shaped from day one. The store contract is what a real ADT/HL7/FHIR feed would push, so the simulation can be replaced without touching the UI.

2. The twelve domains

The platform was scoped across the twelve domains that actually make a hospital run:

  1. Patient Flow & Throughput — ED/urgent-care, triage, admission, transfers, discharge, boarding, LOS
  2. Bed & Space Management — placement, EVS turnover, capacity, isolation/cohorting, surge
  3. Workforce & Staffing — scheduling, sick-call backfill, float pool, ratios, fatigue
  4. Clinical Care Coordination — care teams, consults, results, deterioration, rapid response, handoffs
  5. Perioperative & Procedural — OR scheduling, block utilization, PACU flow, case readiness
  6. Diagnostics & Ancillary — lab/imaging/pharmacy turnaround, transport
  7. Supply, Equipment & Assets — RTLS tracking, PAR resupply, crash carts, biomed, SPD
  8. Patient & Family Experience — wait apps, leave-and-return, family updates, check-in, CAHPS
  9. Safety, Quality & Compliance — infection control, risk scoring, bundles, survey readiness
  10. Command Center, ML & Optimization — forecasting, the optimization brain, digital twin
  11. Integrations & Interoperability — EHR, HL7/FHIR, ADT, nurse call, RTLS, devices
  12. Revenue Cycle & Administrative Ops — prior-auth, utilization mgmt, denials prevention

3. The module catalog (27)

P0 = core command-center, built or central now · P1 = next · P2 = later. ★ = built as a flagship in this repo.

Command Center & Optimization Brain

  • ★ Mission Control Command Wall P0 — single role-aware operating picture fusing census forecast, capacity/surge index, ranked action queue, and live KPIs.
  • ★ Unified Alert, Escalation & Command Feed P0 — one closed-loop stream of every recommendation and alert with owner routing, SLA timers, ack, and auto-escalation ladders.
  • Census & Multi-Horizon Demand Forecasting P1 — probabilistic 4/12/24/72h census and net-bed-position (P10/P50/P90) forecasts that pre-trigger staffing, EVS, and discharge actions.
  • Recommendation Outcome Tracking & Digital-Twin Scenario Lab P2 — closed-loop tracking of every recommendation's impact, plus a discrete-event hospital twin for side-by-side what-if surge planning.

Patient Flow & Throughput

  • ★ Predictive ED Wait-Time & Triage Board P0 — per-patient/per-acuity live wait & door-to-provider forecasts, a waiting-room deterioration watchlist, threshold auto-escalation.
  • Patient-Flow & Surge Orchestration Engine P0 — live ED-to-discharge throughput with boarding/inpatient-demand forecasting and one coordinated surge-response playbook.
  • Discharge Planning & Barrier Orchestrator P1 — live per-patient discharge plan with predicted date, auto-detected barriers routed to owners, discharge-by-noon tracker.

Bed & Space Management

  • ★ Intelligent Bed Placement & Census SSOT P0 — constraint-aware ranked bed matching with one-tap assign, reconciled real-time bed board, infection-control placement guardrails.
  • ★ EVS Turnover & Patient Transport Dispatch P0 — auto-created, demand-prioritized cleaning and transport jobs with smart dispatch, ETAs, and stall escalation that closes the bed-turn loop.

Workforce & Staffing

  • ★ Sick-Call Backfill & Open-Shift Cascade P0 — detects coverage gaps and runs a cost-and-fit-ranked, timed open-shift cascade to fill with one tap before agency is ever called.
  • ★ Ratio, Acuity & Fatigue Compliance Monitor P0 — continuous acuity-weighted ratio surveillance with breach-risk forecasting, fatigue/OT-equity guardrails, immutable compliance log.
  • Float Pool & Cross-Unit Resource Optimizer P1 — house-wide competency-aware float allocation proposing specific rebalancing moves against real-time per-unit demand.

Clinical Care Coordination

  • Closed-Loop Care Coordination & Open-Loops Tracker P1 — every consult, critical result, and order becomes an owned, SLA-timed work item with automatic escalation instead of phone-tag.
  • Deterioration & Sepsis Early-Warning + RRT Activation P1 — continuous composite deterioration/sepsis scoring with tiered alerts and one-tap closed-loop rapid-response activation.

Perioperative & Procedural

  • Perioperative Day-of Throughput Control Tower P1 — first-case readiness, dynamic case re-sequencing, OR turnover choreography, PACU bottleneck management.
  • Block Utilization & Auto-Release Engine P2 — forecasts underused OR blocks and auto-releases reclaimed time to a demand-ranked open-time marketplace.

Diagnostics & Ancillary Services

  • Diagnostics TAT Sentinel & Ancillary Bottleneck Forecast P1 — SLA-clocked lab/imaging/pharmacy turnaround with critical-value closed-loop notification and predictive congestion alerts.
  • Cross-Service Discharge Dependency Resolver P2 — consolidated "what's blocking this patient from leaving" view across lab, imaging, pharmacy, transport with one-tap expedite.

Supply Chain, Equipment & Assets

  • Asset Locator, PAR Resupply & Crash-Cart Readiness P2 — RTLS nearest-available equipment dispatch, autonomous PAR replenishment, live crash-cart/code-readiness assurance.
  • SPD Tray-to-OR & Biomed Predictive Readiness P2 — sterile-tray reprocessing sequenced to the surgical schedule, predictive biomed failure dispatch, recall quarantine.

Patient & Family Experience

  • Patient & Family Status / Wait Experience Layer P1 — live queue position, geofenced leave-and-return, HIPAA-safe perioperative/inpatient milestone updates to patients and families.
  • Digital Check-in, Eligibility & CAHPS Watchtower P2 — touchless pre-registration with real-time eligibility pre-clearance and in-stay sentiment pulses that trigger live service recovery.

Safety, Quality & Compliance

  • Safety & Quality Surveillance Suite P1 — real-time HAI/sepsis/fall/pressure-injury/med-safety surveillance with bundle-compliance nudging and continuous survey-readiness posture.
  • Incident & Risk Management with Trend Detection P2 — low-friction auto-context incident capture, severity auto-triage, cluster/trend detection to surface systemic risk before sentinel events.

Revenue Cycle & Administrative Ops

  • Prior-Auth, Level-of-Care & Denials Prevention Pipeline P1 — concurrent prior-auth orchestration, observation-vs-inpatient status optimization, predictive pre-bill denials prevention.
  • Throughput-to-Margin Financial Control Tower P2 — ties live throughput to contribution margin, revenue-at-risk, and avoidable-day cost so ops decisions are steered by dollars.

Integrations & Interoperability

  • Canonical Event Bus & Interoperability Truth Layer P1 — bidirectional, ADT-triggered event bus with interface-health prediction and RTLS-fused location reconciliation underpinning every module.

4. Flagship deep-dive (built in this repo)

★ Mission Control (/command)

The default home wall and the only place the brain becomes fully visible. Census ribbon of house-wide KPIs with trends; a capacity/surge gauge with a 4/12h boarding forecast; the Action Queue of ranked, pre-justified recommendations (one-tap accept / override); the Watchtower of anomaly & threshold detectors that deep-link into the owning module; the closed-loop impact ledger (boarding hours avoided, moves & agency shifts saved); and the embedded command feed. Switching scenario from the ribbon visibly moves the gauge and queue.

★ ER & Patient Flow (/flow)

A charge-nurse throughput board. A sortable tracking board of every ED patient with live wait timers, predicted door-to-provider (p50/p90), admit probability, and a deterioration dot; a per-acuity wait-forecast tile vs SLA; a deterioration watchlist with trajectory sparklines and one-tap RRT escalation; an ED-boarding forecast with surge band; and a patient drawer exposing the full forecast and the admit-probability score breakdown.

★ Beds & EVS Turnover (/beds)

Single source of truth for ~186 beds. A live bed grid grouped by unit with status colors, attribute icons, and turnover timers; a placement queue where each boarder gets the brain's top-ranked clean bed with a "why this bed" breakdown and one-tap assign; a score drawer with ranked alternatives (override-able); an EVS Kanban (Requested → Assigned → Cleaning → Ready) with ETAs and stall escalation; and a discrepancy reconciler. Accepting a placement fans out an EVS job + transport + a receiving-unit alert into the feed.

★ Staffing Sick-Call Backfill (/staffing)

A fill board of open coverage gaps, each running a tiered cascade (same-unit off-shift → float pool → other-unit OT → agency last) ranked by a transparent accept-likelihood model (fatigue, hours, competency, credential validity, cost); a candidate drawer showing "why this nurse first"; live acuity-weighted ratio gauges per unit with breach detection; a sortable roster with fatigue and credential-expiry flags; and an immutable compliance log.

5. The optimization brain

The brain is intentionally transparent, not a black box: each predictor is an explainable weighted sum of normalized features that ships its breakdown. This is what makes one-tap acceptance safe and auditable.

  • Predictors — admit probability, quantile wait forecast (p50/p90), predicted LOS & discharge, deterioration trajectory, staffer accept-likelihood.
  • Solvers — constraint-aware bed placement (hard: isolation/telemetry/gender/service; soft: ratio balance, LOS fit, minimize moves), tiered backfill cascade ranker, capacity/surge index, boarding forecast.
  • Orchestration — detect → recommend (ranked, justified) → route (owner + SLA + escalation ladder), and on accept, fan out follow-on events (placement ⇒ EVS + transport + receiving alert).

6. Beyond the catalog — high-leverage additional ideas

These are the differentiated, defensible ideas that go past the obvious operational lens. They are not yet built, but the architecture is designed so each slots into the same event bus and brain.

  • Behavioral-Health & Psych-Boarding Orchestrator — a distinct flow lane for the worst real ED boarding pathology: external psych-bed search, involuntary-hold legal clocks, elopement/suicide risk surveillance, sitter-demand optimization, and code-grey/security dispatch. Needs its own placement constraints (ligature-safe rooms, line-of-sight) the bed optimizer should learn.
  • Degraded-Mode / Downtime Command — a first-class business-continuity posture for when feeds go dark or ransomware hits: the wall switches to a downtime view, drives paper-fallback reconciliation, and orchestrates manual catch-up on recovery. A command center that goes blind during the highest-stakes hour is itself a safety risk; resilience is a differentiator.
  • ML Fairness & Override Governance — subgroup calibration dashboards, override-rate as a model-health signal, and explicit autonomy boundaries (which actions may ever auto-execute vs. always human-gated). For a system that allocates scarce beds, shifts, and discharges, provable non-discrimination is both an ethical necessity and a procurement edge.
  • Regional Transfer Center (mission-control-of-mission-controls) — load-balance patients across a health system: system-wide bed availability, EMS destination steering, outside-hospital transfer triage, repatriation. Reuses the same forecasting/optimization brain across nodes.
  • Care-Transitions & Readmission Prevention — turn the 30-day-readmission KPI into an owned workflow: risk-driven interventions, PCP/SNF/home-health handoff, follow-up + med-adherence + remote-monitoring coordination. The most expensive failure mode, currently only measured.
  • HICS / Mass-Casualty Incident Command — a distinct command posture from ordinary surge: START triage, decon/lockdown/evacuation orchestration, external-event feeds. Satisfies CMS Emergency Preparedness and showcases the platform at its highest stakes.
  • Care-delay & equity services as dispatchable — extend the proven "EVS-as-a-dispatchable-service" pattern to interpreter/language access, dietary, social work/SDOH, chaplaincy, and ambient/voice documentation — the barriers an ops-only lens leaves invisible.
  • Adoption & change-management layer — shadow-mode rollout, frontline threshold-tuning, model clinical-validation sign-off, and labor/union considerations for AI-driven staffing and fatigue scoring. These systems die from adoption failure, not algorithmic failure.

7. Known gaps & responsible-AI guardrails

A faithful concept names what it does not yet handle. A real build must address:

  • Service lines missing from the model — behavioral health/psych boarding; pediatrics, NICU, and L&D/obstetrics (couplet care, non-deferrable births) need their own constraints and forecasts.
  • Bias & fairness of the models themselves — triage acuity, deterioration scoring, no-show, denial-risk, and discharge models all carry documented demographic-bias risk (e.g. pulse-ox by skin tone). A dedicated fairness-audit surface and subgroup calibration are first-class, not a footnote.
  • Automation accountability — liability when an auto-action causes harm, automation complacency (rubber-stamped one-tap accepts), override-rate monitoring, and a clear human-authority boundary.
  • Downtime & cyber-resilience — the single most consequential modern hospital scenario; needs a downtime playbook and graceful-degradation UX, not just interface-health prediction.
  • Cross-facility scale, mass-casualty mode, post-acute loop — see §6.
  • Consent & surveillance ethics — RTLS location, hand-hygiene sensors, fatigue scoring, and family geofencing are pervasive monitoring; they need consent management, data minimization, and staff/labor consent for performance tracking.
  • ROI attribution rigor & TCO — every EHR interface is a real integration project; avoided boarding hours must be proven against a baseline, not secular trend; vendor displacement is real.

8. KPIs the platform moves

ED LWBS & door-to-provider · ED boarding hours · inpatient LOS & avoidable days · bed-turn time · discharge-by-noon % · occupancy & capacity index · nurse-ratio compliance · agency/OT spend · first-case on-time starts & OR turnover · diagnostic turnaround · HAI/sepsis-bundle compliance · denial rate · patient experience (CAHPS) · 30-day readmissions.

9. Integration surface (for "going live")

EHR (Epic/Cerner) via HL7v2 & FHIR, ADT event feeds, eligibility (270/271), nurse call, RTLS, secure messaging/paging, EVS/transport dispatch systems, monitor & medical-device integration, scheduling/OR systems, and payer prior-auth APIs — all funneled through the Canonical Event Bus so every module consumes one normalized stream.