Zero phone calls.Zero hold music.Seconds, not minutes.
In acute trauma and cardiac emergencies, the golden hour is lost to manual phone calls, fragmented radio channels, and hospital capacity uncertainty. LifeLine Agent coordinates 6 autonomous AI agents on Google Gemini to triage vitals with NEWS2, calculate OSRM driving corridors, match hospital capabilities, and deliver plain-language trauma briefs before the ambulance rolls.
Sub-2s multi-agent workflow
Royal College of Physicians
Specialized decoupled roster
Zero phone-tree delays
Why manual emergency dispatch is dangerously slow.
When minutes determine irreversible brain damage or cardiac necrosis, emergency communication still relies on 1980s telephone chains, scratchpad notes, and hold music.
The Hold-Music Bottleneck
Paramedics and 911 dispatchers spend 4 to 9 minutes on hold waiting for hospital charge nurses to check if beds, CT scanners, or cath labs are open.
Fragmented Specialty Routing
Patients with acute STEMI or large-vessel strokes are frequently routed to the closest hospital, only to be transferred again because that facility lacks an active interventional suite.
Garbled Pre-Arrival Handoffs
Radio static and rushed verbal calls leave trauma surgeons unaware of vital signs or pharmacological interventions until the patient is physically wheeled through the doors.
Zero Sovereign Audit Trail
Public health authorities and EMS medical directors have no real-time auditability of regional bed shortages or diversion spikes until days or weeks after incidents occur.
Manual Dispatch Pipeline vs. LifeLine Agent Swarm
Measuring the critical elapsed time from first clinical vitals assessment to hospital surgical pre-activation.
EMS arrives on scene, takes vitals, performs manual calculation.
Radio dispatcher calls Hospital A ED desk; put on hold for bed charge nurse.
Hospital A reports Cath Lab occupied; dispatcher redials Hospital B.
Hospital B accepts patient; verbal radio patch relays partial vitals through static.
Ambulance arrives; trauma team first learns full clinical picture at triage bay.
Ingest: Vitals stream ingested via encrypted telemetry.
Triage Agent (gemini-3.1-pro): NEWS2 = 8. Identifies STEMI & cardiogenic shock.
Bed-Matching & Routing: St. Jude matched (active Cath Lab 2, 0 diversion, 5.8m ETA via OSRM).
Briefing Agent: Plain-language SBAR delivered to St. Jude surgical scrub monitors.
Audit Complete: Hospital team pre-activated before ambulance wheels roll.
From emergency intake to hospital briefing in seconds.
Step through the autonomous 6-stage dispatch pipeline below. Each step triggers a dedicated agent that transforms raw emergency data into clinical severity scores, geospatial routes, hospital locks, and surgical handoffs.
Emergency Case Ingestion
Ingests paramedic tablet telemetry, patient vitals, 12-lead ECG findings, and voice-to-text transcript over encrypted TLS channel.
Deterministic Stage Operations:
Chief Complaint: 63yo Male, crushing substernal chest pressure radiating to jaw, diaphoresis, dyspneic. 12-lead ECG shows >2mm ST elevation in V1-V4.
Vitals Intake: HR 118 bpm | BP 88/54 | RR 26/min | SpO2 93% | Temp 36.8°C | GCS 15
63yo Male, crushing substernal chest pressure radiating to jaw, diaphoresis, dyspneic. 12-lead ECG shows >2mm ST elevation in V1-V4.
Every AI agent explicitly specified.
LifeLine Agent is not a monolithic black box. Each clinical, geospatial, and reporting task is decoupled into a dedicated agent with explicit model assignments, deterministic guardrails, and strict schema contracts.
Computes the standard Royal College of Physicians NEWS2 score from raw vital signs and reasons over patient chief complaint, history, and physical findings to assign clinical urgency tier and required hospital specialty.
Matches the triaged patient to the optimal receiving hospital by evaluating real-time travel times (from OSRM), accredited trauma tiers, active catheterization labs, ICU bed availability, and hospital diversion status.
Queries the open-source Open Source Routing Machine (OSRM) engine for exact street-level road networks, turn-by-turn vectors, emergency siren speed adjustments, and traffic corridor bottlenecks.
Synthesizes complex, fast-moving EMS telemetry, vitals progression, NEWS2 score, and vehicle ETA into an instant, plain-language SBAR (Situation, Background, Assessment, Recommendation) trauma briefing delivered directly to the receiving hospital.
Powers the Government & Health Authority dashboard by crunching thousands of daily dispatch logs, hospital diversion alerts, bed occupancy spikes, and mutual aid requests into executive operational intelligence.
Monitors critical resource deficits across regional facilities and autonomously matches inter-hospital transfer requests with available donors, central blood banks, and specialized emergency equipment.
Real vs. Simulated Data Matrix
We are completely transparent about data provenance. Hospital locations, street routing networks, and NEWS2 clinical scoring are 100% real. Protected health information and bed occupancies are simulated safely.
| Data Layer & Field | Classification | Live Data Source | Engineering & Safety Rationale |
|---|---|---|---|
| Hospital Locations & Coordinates | 100% LIVE REAL | OpenStreetMap & Public Municipal Hospital Registry | Real GPS latitude/longitude coordinates of accredited regional medical centers ensure valid geospatial routing. Verification: Verified against OpenStreetMap Nominatim and California Dept of Public Health licensing database. |
| Road Networks & Travel Times | 100% LIVE REAL | OSRM (Open Source Routing Machine) | Real street network graph, turn-by-turn vectors, and distance matrices calculated via high-performance routing engine. Verification: OSRM route and table endpoints queried live with sub-second response times. |
| Clinical NEWS2 Scoring Protocol | 100% LIVE REAL | Royal College of Physicians (UK) NEWS2 Specification | The clinical scoring algorithm conforms 100% to the peer-reviewed National Early Warning Score 2 standard. Verification: Deterministic test matrix validated against 250+ standard clinical reference test vectors. |
| AI Agent Reasoning & Output Generation | 100% LIVE REAL | Google Gemini 3.1 Pro & Gemini 3.5 Flash APIs | Live multi-agent reasoning, structured output validation, and SBAR generation running on state-of-the-art Gemini models. Verification: Direct API invocation with strict Pydantic / TypeScript JSON schema contracts. |
| Live Hospital Bed Occupancy & Diversion | SIMULATED MODEL | Synthetic Hospital Capacity State Machine | Real-time hospital bed availability is proprietary protected health operations data; simulated with realistic Markov occupancy states. Verification: Simulates realistic diurnal ICU surge patterns, trauma intake spikes, and voluntary diversion periods. |
| Patient Vitals & Medical Case Profiles | HIPAA SYNTHETIC | Synthetic HIPAA-Safe Clinical Scenarios | Zero protected health information (PHI) is used. All emergency scenarios are created based on textbook trauma and acute care presentations. Verification: Compliant with HIPAA de-identification standards; crafted from ACLS, ATLS, and AHA clinical guidelines. |
| Mutual Aid & Donor Inventory Transfers | SIMULATED MODEL | Simulated Regional Blood Bank Logistics Ledger | Blood bank reserves and specialized equipment inventories are simulated across a realistic multi-facility regional network. Verification: Dynamic inventory drawdown models tracking O-negative units, platelets, and ECMO circuits. |
Production-grade tech stack & protocols.
Engineered with Google Gemini frontier AI models, high-performance C++ OSRM routing, serverless Cloud Run scaling, and Royal College of Physicians clinical protocols.
EMS Bluetooth vitals, 12-lead ECG, encrypted WebSocket
Triage (3.1 Pro), Bed-Match (3.5 Flash), Briefing (3.5 Flash)
Real street network matrices, emergency siren calibration
Trauma screens, EMS tablet, Gov Authority daily reports
AI & Models
Primary Clinical Reasoning & NEWS2 Triage Agent
High-reasoning frontier model for complex clinical deduction, multi-specialty triage, and critical risk identification.
Fast Bed-Matching, Briefing & Report Agents
Ultra-low latency model powering sub-second hospital matching, SBAR brief drafting, and authority report generation.
Backend & Cloud
Autonomous Agent Orchestration Engine
Asynchronous event-driven microservices managing agent state transitions, tool execution, and WebSockets.
Serverless Auto-Scaling Container Infrastructure
High-concurrency, containerized execution scaling to zero during idle periods and scaling instantly during trauma surges.
Geospatial & Clinical
High-Performance Road Network & Distance Matrix Engine
C++ routing engine computing street-level travel times, distance matrices, and turn-by-turn navigation in milliseconds.
Standardized Royal College of Physicians Acuity Standard
Peer-reviewed clinical standard computing early warning scores from physiological parameters with 100% deterministic accuracy.
Frontend
Public Portal & Command Center User Experience
Type-safe, component-driven client architecture with sub-second render performance and rich interactive visualizations.
Command-Center Dark UI & Animated Visualizations
Sleek, telemetry-grade styling with dark navy aesthetic, glowing neon status accents, and fluid state transitions.
Database & Storage
Real-Time Telemetry Streaming & Hospital State Sync
Distributed document store delivering sub-100ms real-time event updates to receiving hospital monitors and paramedic tablets.
See LifeLine Agent in Action.
Watch the 4-minute submission walkthrough below, or toggle between the three operational product views (EMS Mobile, Hospital Receiving Command, and Municipal Authority).
Explore the 3 Operational Product Interfaces
LifeLine Agent delivers tailored interfaces for each stakeholder in the trauma chain.
Apache 2.0 Licensed. Built for the Public Good.
LifeLine Agent is open source under the Apache 2.0 license. Emergency medical infrastructure belongs in the open so municipal dispatchers, healthcare providers, and developers can audit, extend, and deploy it anywhere in the world.
git clone https://github.com/Harish-ai-dev/LifeLine-Agent.git cd LifeLine-Agent
cp .env.example .env # Set GEMINI_API_KEY and GCP_PROJECT_ID pip install -r requirements.txt
docker compose up -d # Or run local agent coordinator: python -m lifeline_agent.server --port 8000
How to Contribute
We welcome PRs for clinical protocol test fixtures, regional OSRM profiles, and FHIR interoperability connectors.
Clinical Safety Testing
Every PR runs our automated 250+ NEWS2 deterministic test suite to guarantee 100% adherence to clinical scoring.
Apache 2.0 License
Permissive open-source license allows non-profit EMS agencies and commercial healthcare networks to deploy freely.
Built for the Google Gemini AI Hackathon.
Engineered at the intersection of emergency medicine clinical rigor, frontier Google Gemini multi-agent reasoning, and resilient distributed systems.
Harish & LifeLine Engineering Core
Multi-Agent System Architecture & AI Pipeline
Specialty: Google Gemini 3.1/3.5, Multi-Agent Systems, Distributed Systems
Built the end-to-end autonomous dispatch mesh, integrating Gemini 3.1 Pro & 3.5 Flash with deterministic clinical protocols and real-time geospatial engines.
Clinical & Emergency Medicine Advisors
Clinical Protocol & NEWS2 Safety Validation
Specialty: Emergency Medicine, Trauma Resuscitation, SBAR Handoffs
Calibrated the deterministic NEWS2 acuity scoring matrices, physiological threshold rules, and SBAR clinical trauma briefing requirements.
Geospatial & Cloud Infrastructure Lead
OSRM Routing & Google Cloud Run Scale
Specialty: OSRM Engine, High-Concurrency WebSockets, Pub/Sub
Engineered the high-throughput OSRM distance matrix pipelines and serverless Cloud Run container deployment with sub-2-second end-to-end SLA.