MediSort

How MediSort uses Kurai to power their newest projects.

Industry

Healthcare & HealthTech

Location

Boston, MA

Employees

200

Identity Provider

MediSort

Workloads

ML Model, Healthcare API, FHIR Integration, XGBoost

About

MediSort is a digital health platform serving 150 hospitals and 500,000+ patients annually. Their ER triage system helps medical staff prioritize patients based on symptoms, vitals, and medical history.

Challenge

ER nurses spent 8-12 minutes per patient on manual triage, leading to 45-minute average wait times. During peak hours, critically ill patients waited 20+ minutes for assessment. Human triage was inconsistent and error-prone.

Solution

Kurai built an ML-powered triage system using XGBoost trained on 2M historical patient records. The system predicts patient acuity (ESI levels 1-5) in real-time with 94% accuracy, integrating with Epic EHR via FHIR API.

Results

70% faster triage (12 min → 3.5 min per patient)...

28% reduction in patients leaving without care...

15% improvement in critical patient identification...

$2.8M annual labor cost savings...

ML-Powered Patient Triage: 70% Faster Assessment

The Problem

Emergency rooms were overwhelmed. With 8-12 minutes required per manual triage assessment, patients waited 45+ minutes during peak hours. Critically ill patients (ESI level 1-2) sometimes waited 20+ minutes for assessment—a life-threatening delay.

Dr. Sarah Chen, ER Director: “We needed a way to identify critical patients faster, but triage nurses were already working at maximum capacity.”

The Solution

Kurai deployed an ML-powered triage system in 14 weeks using XGBoost gradient boosting models trained on 2M historical patient encounters. The system predicts ESI acuity levels (1=critical, 5=non-urgent) in real-time with 94% accuracy.

Model Architecture:

  • Features: 47 features (symptoms, vitals, age, history, medications)
  • Algorithm: XGBoost with SHAP for explainability
  • Deployment: FastAPI + AWS ECS + Redis caching
  • Integration: Epic FHIR R4 API with HIPAA compliance

The Results

  • Triage time: 12 min → 3.5 min (70% faster)
  • Critical patient wait: 20 min → 4 min (80% reduction)
  • Patients leaving without care: 12% → 8.6% (28% reduction)
  • Labor savings: $2.8M/year
  • Model accuracy: 94% (97% sensitivity for critical patients)

Technology Stack

XGBoost 2.0, FastAPI, Epic FHIR R4 API, PostgreSQL RDS, Redis ElastiCache, AWS ECS Fargate, Prometheus + Grafana monitoring

Key Metrics

MetricBeforeAfterImprovement
Triage time12 min3.5 min-70%
Critical wait20 min4 min-80%
Left without care12%8.6%-28%
AccuracyN/A94%New

MediSort’s triage system now assesses 500K+ patients annually, saving lives by identifying critical patients 80% faster.

Trusted by Industry Leaders

Empowering innovators, shaping the future

David Gutierrez

David Gutierrez

CTO at TechFlow AI

"The RAG system they built for us reduced our support tickets by 60%. Their expertise in LLM integration is unmatched."

Pierluigi Camomillo

Pierluigi Camomillo

VP Engineering at DataScale

"They migrated our monolith to microservices seamlessly. We saw a 40% cost reduction and significantly improved scalability."

Ella Svensson

Ella Svensson

Founder at MediSort Health

"Their ML-powered patient triage system transformed our operations. 70% faster triage with 94% accuracy—sim incredible results."

Alexa Rios

Alexa Rios

Chief Product Officer at ShopMax

"The recommendation engine they built increased our AOV by 32%. Highly recommended for any e-commerce business looking to leverage AI!"