AI-Powered Genomic Report Summarization with AWS

The challenge

RxGenomix clinicians were spending significant time, often several hours, manually reviewing and summarizing 40–60 page genomic and medical reports to create provider-ready Medical Action Plans (MAPs). This labor-intensive workflow posed several critical challenges:

Time-consuming, repetitive work for clinicians
Prone to human error and inconsistencies
High risk of missing gene-drug interactions or recommendations
Limited ability to scale with increasing patient volume and data complexity

To improve operational efficiency and reduce manual effort, the customer needed an automated, secure, and accurate solution that could integrate with existing clinical workflows and meet compliance standards.

Primary objectives

Our customer set out to transform their Medical Action Plan (MAP) generation process with the following goals:

Automation: Eliminate manual summarization to reduce clinician workload
Scalability: Process a higher volume of patient reports without proportional resource increase
Accuracy: Ensure consistent, error-free extraction of complex genomic data
Compliance: Maintain full auditability and data security in alignment with HIPAA standards
Provider Adoption: Deliver AI-generated summaries that providers could trust and refine

Architecture and services used

To meet the outlined objectives, zeb experts designed a robust solution using AWS-native services and AI models from Amazon Bedrock.

Amazon S3
Served as the foundational storage for all incoming genomic reports, including XML files and scanned PDFs. Data was securely stored in encrypted buckets with fine-grained IAM-based access controls.
AWS ECS Fargate
To manage and orchestrate the summarization workflow, RxGenomix utilized Amazon ECS with Fargate. Containerized microservices handled tasks such as ingesting reports from S3, formatting medication and gene interaction data, invoking Claude 3.5 Sonnet via Amazon Bedrock, and generating structured action plans. Fargate provided a fully managed, serverless compute environment for containers, allowing RxGenomix to scale on demand while avoiding the operational complexity of managing servers or provisioning infrastructure manually.
Amazon Bedrock with Claude 3.7 Sonnet
Powered the core summarization engine that interprets large, complex pharmacogenomic reports and converts them into actionable MAPs suitable for provider use.
AWS Secrets Manager
Managed all API credentials and secure endpoints. This eliminated the need for hardcoded secrets and ensured centralized, auditable access to sensitive data and services.
AWS CodeCommit + CodePipeline
To support model iteration and controlled feature releases, CodeCommit and CodePipeline were used to manage application and prompt templates. Each model version passed through a structured approval flow, enabling audit-ready updates in a healthcare environment.
Provider Feedback Loop
MAPs included a built-in feedback mechanism, allowing providers to annotate drafts with suggestions or corrections. These annotations were collected via Lambda APIs and persisted in S3 as structured feedback files, enabling future model fine-tuning and personalization.

Timeline
The project was completed within 3 months through phased iterations, starting with foundational infrastructure and progressively incorporating AI-based summarization and feedback capabilities. Early integration of provider feedback enabled rapid adoption and continuous refinement, leading to measurable impact in a short span of time.

KPIs and outcomes

Operational KPIs

93–95% accuracy in extracting actionable content from long genomic and medical reports
95% reduction in manual review time, allowing clinicians to spend more time on patient care
150% increase in daily processing capacity, enabling scalable report throughput

Clinical & Adoption Outcomes

High provider adoption and trust, driven by the embedded feedback loop
Reduced risk of missed interactions, supporting safer care delivery

Business Impact

Positioned RxGenomix as a frontrunner in scalable pharmacogenomics delivery
Enabled AI-driven efficiency while maintaining human oversight

Lessons learned

Prompt engineering is critical: Success with AI in healthcare hinges on domain-aware prompts and medically accurate templates.
Feedback isn’t optional, it’s strategic: Clinician annotations helped refine AI output and foster trust.
Compliance must be native to design: AWS-native services were chosen specifically for their auditability and HIPAA-aligned capabilities.
Fit-for-purpose automation drives ROI: Embedding AI within existing EMR workflows ensured adoption and sustained impact.

Our customer’s journey to automate Medical Action Plan generation with zeb’s AWS-based solution led to a complete transformation in their clinical operations. By integrating Amazon Bedrock with Claude 3.7 Sonnet into a secure, ECS pipeline, they reduced manual effort by 95% and tripled their daily processing capacity. The new system delivered accurate, compliant, and provider-ready summaries, empowering clinicians to focus more on patient care. This AI-powered automation positioned RxGenomix to scale personalized pharmacogenomic services while maintaining clinical oversight and trust, setting a new benchmark for care delivery efficiency.