About Customer
CareOps AI is a leading home health services organization delivering comprehensive patient care across multiple regions. The organization manages complex patient intake and referral workflows involving clinical staff, administrative personnel, and EHR specialists.
The organization processes hundreds of patient referrals daily, each requiring accurate extraction of structured patient data from multi-page medical document packages. Timely and accurate data extraction is critical to care delivery, payor coordination, and EHR onboarding.
CareOps AI operates across a multi-region healthcare system with 200+ clinical staff, processing high volumes of home health referral documents under strict HIPAA compliance requirements.
Background
A multi-region home health provider operating with 200+ clinical staff engaged CAREOPS.AI to automate its patient intake process. The organization processes hundreds of patient referrals daily, each accompanied by multi-page medical document packages — clinical notes, physician orders, insurance information, medication lists, and patient history. Accurate, timely extraction of structured data from these documents is critical to initiating care, coordinating with payors, and onboarding patients into the EHR system.
Challenges
CareOps AI is a leading home health services organization delivering comprehensive patient care across multiple regions. The organization manages complex patient intake and referral workflows involving clinical staff, administrative personnel, and EHR specialists.
CareOps AI processes hundreds of patient referrals daily, each accompanied by multi-page medical document packages — including clinical notes, physician orders, insurance information, medication lists, and patient history records. Efficient and accurate extraction of structured patient data from these documents is critical to initiating timely care delivery, coordinating with payors, and onboarding patients into their Electronic Health Record (EHR) system.
- Time-Consuming Manual Data Entry: Clinical and administrative staff manually extracted patient details (demographics, insurance identifiers, physician contacts, medication lists, diagnosis codes) from multi-page PDF referral packages and entered them into the EHR system, consuming significant staff hours per referral.
- High Error Rate: Manual transcription frequently resulted in incorrect medication names, insurance identifiers, physician contact details, and diagnosis codes — causing downstream inconsistencies across patient records.
- Multiple Document Formats: Referral packages arrived as multi-page PDFs containing mixed content types — physician orders, clinical assessments, insurance forms, and medication records — with no standardized layout, making manual extraction inconsistent.
- Scalability Constraints: The manual process could not scale with growing referral volumes. Handling admission spikes required temporary staff, extended hours, and significant coordination overhead — without guaranteeing quality or turnaround time.
- Slow Processing Time: End-to-end processing of a single referral package — from document receipt to structured data entry — took hours, delaying downstream workflows including insurance verification, physician coordination, and care scheduling.
Compliance and Audit Risks: Inconsistent extraction processes created gaps in audit trails. Without a standardized, traceable extraction pipeline, demonstrating HIPAA-compliant data handling was difficult and resource-intensive.
Solution
CalyzaTech has designed and deployed Nexus-AI, an AWS-native, cloud-native medical document intelligence platform. This platform automatically transforms multi-PDF referral packages into fully structured, schema-validated JSON records at scale.
Nexus-AI combines Amazon Textract for OCR-based text extraction with Amazon Bedrock (utilizing Amazon Nova Pro) for intelligent, structured data extraction. The system is orchestrated through an event-driven architecture that features comprehensive monitoring and automated notifications.
The Nexus-AI pipeline processes each referral through a strict 7-stage automated workflow:
- Ingestion & Validation: Receives the API request and performs nine integrity validation checks before processing.
- Job Tracking: Generates a unique Job ID, sets job status to “In_Progress” in AWS Systems Manager, and initiates structured logging in AWS CloudWatch.
- Smart OCR Caching: Utilizes Amazon Textract via parallel workers for digitization. The system checks Amazon S3 for an existing OCR output file before processing; if found, the cached result is reused, skipping Textract entirely and reducing processing time by up to 75%.
- Context Aggregation: Consolidates raw Textract outputs into a single, unified context document for FM reasoning.
- Multi-Section FM Extraction: Amazon Bedrock (Amazon Nova Pro) sequentially and contextually extracts 8 medical record sections (e.g., patientProfile, medicationsAndAllergies) via section-specific prompts.FM output is validated, programmatically repaired if required, and normalized against a strict JSON schema.
- Structured Storage: The complete canonical JSON record and individual section JSON files are saved securely in the patient’s Amazon S3 folder.
- Handoff & Notification: The SSM job status is updated to Completed. A structured success message containing the workflow summary and S3 output paths is sent downstream via Amazon SQS for asynchronous ingestion into the CareOps platform.
AWS Services
The platform runs on a HIPAA-aligned, cloud-native architecture with encrypted storage, secure secrets management, containerized and horizontally scalable compute, and full observability.
Solution
CAREOPS.AI deployed its cloud-native medical document intelligence platform to automate the provider’s end-to-end intake workflow, transforming multi-page PDF referral packages into fully structured, schema-validated patient records at scale.
At its core, the platform combines OCR-based text extraction with LLM-powered structured data extraction, orchestrated through a secure, event-driven cloud architecture with built-in monitoring and automated notifications. Each referral moves through an automated pipeline that:
- Ingests and validates each referral package on receipt
- Digitizes the documents using OCR
- Extracts and structures the key medical record sections using LLM-based reasoning
- Validates and normalizes every field against a strict data schema
- Securely stores the structured records and hands them off downstream for ingestion into the CAREOPS.AI platform, with full job tracking and audit logging throughout
Outcome
- Efficiency: Automated workflows process complex referrals in ~6–7 minutes (Amazon Textract OCR in <10 seconds), replacing hours of manual review and significantly reducing data entry.
- Accuracy: Amazon Textract (OCR) and Amazon Bedrock (FM extraction) generate high-accuracy, schema-validated JSON across all canonical sections.
- Performance: Containerized deployment on Amazon ECS Fargate combined with parallel architecture scales horizontally to handle volume spikes and concurrent packages seamlessly.
- Observability: Real-time job state tracking via AWS Systems Manager and structured, job-correlated telemetry in AWS CloudWatch provide full observability.
- Compliance: HIPAA-aligned with private, encrypted Amazon S3 storage, no PII in logs, and secure key management via AWS Secrets Manager.
- Cost Optimization: Smart S3 caching reduces repeat Amazon Textract costs by up to 75%, while serverless compute (Fargate) maintains a strictly pay-per-use model.