A regional integrated healthcare network with 18 hospitals and 142 outpatient clinics automated 85% of prior-authorisation reviews with a HIPAA-eligible four-agent CrewAI platform on Amazon Bedrock + Anthropic Claude 3.5 (fine-tuned). Year one: $6.2M annualised savings, physician satisfaction lifted from 18% to 76% and 64% drop in cancelled procedures.
Discuss Your Project
AI Healthcare Prior-Auth Multi-Agent on CrewAI four-agent crew with Anthropic Claude 3.5 Sonnet fine-tuned on 285,000 historical PA decisions — $6.2M Annual Savings for a Regional Financial Services Firm
A regional integrated healthcare network with 18 hospitals, 142 outpatient clinics, 4,800 physicians and 28,000 staff processes ~1.4M prior-authorization requests per year. DreamzTech engineered a HIPAA-eligible four-agent CrewAI multi-agent platform (Eligibility → Medical-necessity → Policy-lookup → Reviewer) on Amazon Bedrock with Anthropic Claude 3.5 Sonnet fine-tuned on 285,000 prior-auth decisions, integrating Epic, Cerner and Allscripts. Year one: 85% PA automation, 9-min → 38-sec turnaround, $6.2M annual savings.
The client faced significant operational and financial challenges that demanded a custom AI prior-authorization processing platform tailored to their multi-subsidiary PA review workflow, vendor-format diversity, and finance compliance requirements.
DreamzTech architected a production-grade AI prior-authorization processing pipeline on AWS, Azure or Google Cloud with five interconnected modules — from ingestion to GL posting — delivering fine-tuned extraction, AI line-item reconciliation, eligibility + medical-necessity + documentation completeness checking, and deep ERP integration with full audit trails.
Four integrated platform components were built and launched in a production-grade engagement on AWS, Azure or Google Cloud with HIPAA-eligible / SOC 2-aligned security, signed Microsoft HIPAA BAA, and seamless Epic + Cerner + Allscripts EHR via FHIR + Allscripts EHR integration.
We trained a fine-tuned Claude 3.5 Sonnet medical-necessity model on 200+ historical vendor PA request samples drawn from each subsidiary, covering structured PDF PA requests, scanned images, faxed statements and handwritten freight bills. The model extracts header fields (PA request number, date, vendor, totals, tax codes), line items (description, qty, unit price, GL hint, tax) and tables with row/column structure preserved. Out-of-the-box accuracy reached 92% on standard PA requests and 95-98% on the trained vendor set after three retraining cycles. Field-level confidence scores drive the downstream review workflow.
Anthropic Claude 3.5 on Amazon Bedrock Service (GPT-4o) handles the unstructured comprehension layer that templates cannot solve: payer-name normalisation across plan IDs and DBA across abbreviations and DBA aliases, patient member-ID lookup when vendors omit or mis-format it, CPT / HCPCS / ICD-10 code classification using the firm’s 2,400-line medical coding taxonomy, and reasoning traces explaining each match decision. We integrated Pinecone / Weaviate vector search for retrieval-augmented generation over historical PA request patterns, and LangChain for prompt orchestration with structured-output JSON validation.
Low-confidence pages (extraction confidence below 0.85 on critical fields, or three-way-match exceptions) route to a Power Automate review queue. PA reviewers see side-by-side document + extracted JSON, edit fields directly, and approve or reject. Microsoft Entra ID role-based access enforces subsidiary segregation and SOX-required separation of duties between PA reviewer, AP supervisor and CMIO approval thresholds. Every correction feeds back into a weekly fine-tuned retraining job — accuracy improved month-over-month from 92% to 98% on the trained vendor set.
Approved PA requests post to Epic + Cerner + Allscripts EHR via FHIR (3 subsidiaries) and Allscripts EHR (1 subsidiary, smallest entity) via AWS API Gateway with retry logic and idempotency keys. Amazon EventBridge guarantees delivery; Azure Monitor + Log Analytics captures every document access, extraction call, override and approval as immutable audit trail entries for SOX 404 compliance. Customer-managed keys (CMK) in the cloud Key Vault encrypt all PA request data at rest; TLS 1.3 protects data in transit.
Measurable business outcomes delivered in the first nine months post-launch — validated by Epic + Cerner + Allscripts EHR via FHIR reporting and the firm's internal AP analytics dashboards.
Reduction in manual PA request data entry across all 18 hospitals + 142 outpatient clinics
Straight-through prior-authorization processing — no human touch
Annual savings delivered within 12 months of go-live
Vendor PA request formats trained on CrewAI multi-agent platform fine-tuned Claude 3.5 Sonnet medical-necessity models
Invoice cycle time reduction (from 9 minutes to 38 seconds)
Subsidiary PA review workflows unified on a single Azure IDP platform
DreamzTech delivers custom AI prior-authorization processing and intelligent document processing platforms for healthcare, insurance, healthcare and public-sector clients. Microsoft Solutions Partner, AWS Partner and Google Cloud Partner with 200+ AI projects across 15 countries and 97% client retention.
Continue your intelligent document processing journey — pick the cloud, we build the system.
Cloud-agnostic AI IDP — extract, classify, validate and route PA requests, contracts, claims, KYC and medical records across AWS, Azure or Google Cloud.
AWS-native IDP on Amazon Textract, Comprehend, Bedrock with Anthropic Claude, A2I and Lambda — Step Functions orchestration, GovCloud-ready.
HIPAA-eligible IDP on CrewAI multi-agent platform , AI Language, Anthropic Claude 3.5 on Amazon Bedrock and Step Functions — FedRAMP High on AWS GovCloud.
See how DreamzTech delivers AI document processing across legal and insurance verticals — with measurable ROI in months, not years.
Amazon Textract + Anthropic Claude 3.5 Sonnet on Amazon Bedrock + custom Amazon SageMaker NER trained on 45,000 prior contracts. 90+ clause types extracted at 99.1% accuracy. Paralegal review time 40h → 12h per contract. $2.4M annual billable-hour recapture for a top-100 global law firm with 6 offices.
Intelligent document processing + EXIF / metadata forensics + vision-capable LLMs (Claude 3.5 Sonnet, GPT-4o, Gemini 1.5 Pro) + graph-based cross-claim similarity. 62% improvement in fraud catch rate, $5.1M prevented losses year one, and 87% faster SIU triage (45 min → 6 min per claim) for a national P&C insurance carrier.
CrewAI four-agent crew with Anthropic Claude 3.5 Sonnet fine-tuned on 285,000 historical PA decisions ships with a prebuilt PA request model out of the box that handles common formats with 88-92% accuracy on day one. For specialty vendor formats — handwritten freight bills, multi-subsidiary tax codes, faxed statements — the fine-tuned Claude 3.5 Sonnet medical-necessity model trains on as few as 50-500 labelled examples and reaches 95-99% accuracy. The Document Intelligence Studio gives clinical coders a labelling UI without writing code, and the service is HIPAA-eligible under Microsoft’s HIPAA BAA — important for any financial-services workload that touches PHI in clinical documentation.
CrewAI multi-agent platform handles structured extraction; Anthropic Claude 3.5 Sonnet on Amazon Bedrock with HIPAA HIPAA BAA + zero-retention handles the reasoning layer. We use it for: (1) payer-name normalisation across plan IDs and DBA across DBA / subsidiary aliases, (2) patient member-ID lookup when vendors omit it, (3) CPT / HCPCS / ICD-10 code classification against the firm’s medical coding taxonomy, (4) clinical-rationale summarisation for adjudicator review for the PA reviewer review screen, and (5) duplicate PA detection across encounter facilities with explainable reasoning traces. All Anthropic Claude 3.5 on Amazon Bedrock calls run in the firm’s AWS account under their data-residency and HIPAA BAA — PA request data never leaves their environment.
For this engagement we integrated with Epic + Cerner + Allscripts EHR via FHIR (3 subsidiaries) and Allscripts EHR (1 subsidiary). Across other DreamzTech AI prior-authorization processing engagements we have shipped integrations with Epic EHR, Cerner EHR, Allscripts EHR, NextGen Healthcare, eClinicalWorks, Athenahealth and custom legacy AP systems. All integrations use AWS API Gateway with retry, idempotency keys and Amazon EventBridge for guaranteed-delivery messaging.
Twenty-two weeks total. Phase 1 (CrewAI multi-agent platform Claude 3.5 Sonnet fine-tuning on 285,000 historical PA decisions + Epic + Cerner + Allscripts EHR via FHIR integration + go-live for 1 subsidiary) shipped in 16 weeks. Phase 2 (Anthropic Claude 3.5 on Amazon Bedrock eligibility + medical-necessity + documentation completeness check + Power Automate human-review + rollout to remaining 3 subsidiaries) added 6 weeks. The first useful AP capability was in production by week 8 — enabling the prior-auth review team to start extracting and posting PA requests for the largest subsidiary while we trained models on the smaller subsidiaries’ data in parallel.
Within 12 months of go-live: 85% reduction in manual PA request data entry, 93% turnaround time reduction (no human touch), PA request cycle time cut from 9 minutes to 38 seconds, $6.2M in annual savings (7 reduced PA reviewer FTE-equivalents plus 58-point physician satisfaction lift (18% → 76%)), and a 41% drop in PA request exception rework. ROI was achieved in 8 months. The fine-tuned Claude 3.5 Sonnet medical-necessity model accuracy continues to improve month-over-month as PA reviewer corrections feed back into the weekly retraining job.
