HealthTech ISV Moves Patient-Data AI from Two Clouds to One Without Losing a Safety Filter, with OpenAI on Amazon Bedrock
At a glance
A HealthTech software company runs its application on AWS but sent every patient prompt to a second cloud, because Azure was once the only place to get OpenAI models with enterprise guardrails. Two cloud vendors processed patient data, and the company had no direct line to OpenAI.
When OpenAI models arrived on Amazon Bedrock, New Math Data moved the workload in six weeks with the same GPT models. Every Azure content filter was rebuilt and tested in Bedrock Guardrails, including two with no direct equivalent, and no partner saw a patient prompt.
Patient data now runs through one cloud vendor under the existing AWS BAA, and the company works directly with OpenAI engineers and AWS architects across 600 clinics. New OpenAI capabilities reach it as they launch on AWS, and its Bedrock spend of about $160K a month counts toward its AWS commitment.
Industry
Use Case
AI Platform Migration & Implementation, Compliance & Governance, Responsible AI, Security & Monitoring
Solution implemented
- OpenAI GPT models on Amazon Bedrock replace Azure endpoints, authentication, and deployment names behind the company's existing gateway.
- Bedrock Guardrails match every Foundry filter, with self-harm and protected-material screening rebuilt as denied topics and word filters.
- Patient identifiers are masked on inputs and outputs, with CloudTrail logging and zero-operator-access controls applied.
- Shadow mode in the company's account mirrors live requests to Bedrock before any production traffic moves.
- Feature-by-feature cutover keeps rollback to Azure until each clinical workflow matches on evaluation.
- IAM roles and PrivateLink replace API keys, and Terraform deploys through the company's existing pipelines.
The value equation
- Patient data runs through one cloud vendor under the existing AWS BAA, simplifying every security review.
- Every Azure content filter carries over and no partner sees patient prompts, so cutover adds no exposure.
- A direct OpenAI relationship brings applied AI engineers and AWS solution architects to the same account.
- The company gets new OpenAI capabilities as they launch on AWS, with no Azure commitment to unwind.
- Model spend of about $160K a month now counts toward the company's AWS commitment.
Company Snapshot
A HealthTech ISV builds clinical documentation and prior-authorization software for outpatient clinics, and now runs its OpenAI-powered AI on the same AWS cloud as its application across 600 clinics.
Location
United States
Customer Situation
A HealthTech ISV sells clinical documentation and prior-authorization software to outpatient clinics. Its application, data, and identity run on AWS, but its AI ran on Azure AI Foundry because that was the only place to get OpenAI models with enterprise content filters. Every patient prompt crossed to a second cloud.
Two cloud vendors processed patient data, which complicated every security review. Microsoft held the OpenAI relationship, so the company had no direct line to OpenAI engineers. Leadership set three conditions for any move: no partner sees patient prompts, no loss of safety filters, and no engineers pulled off a launch already underway.
Many HealthTech ISVs built AI on Azure before OpenAI models reached AWS and carry the same two-cloud split today.
NMD Solution
NMD reviewed and found that both reasons the company chose Azure, OpenAI models and enterprise guardrails, now existed on Bedrock, and that the move would give the company a direct relationship with OpenAI. The solution required a provider layer behind the existing gateway, every Foundry filter mapped to Bedrock Guardrails and calibrated on a known-good and known-bad test set, and shadow mode and evaluation inside the company’s account so NMD never sees a prompt. Terraform and IAM roles deploy it through existing pipelines. One company engineer reviewed pull requests about four hours a week. Clinical leads scored outputs from both clouds before any traffic moved.
What We Delivered
OpenAI runs on Amazon Bedrock in production for all 600 clinics, six weeks after kickoff, with the same GPT models the company used on Azure. Every content filter carries over and was proven against a test set before cutover. No partner sees patient prompts. Patient data runs through one cloud vendor under the existing AWS BAA, and the company works with OpenAI applied AI engineers and AWS architects on the same account. Clinical leads scored outputs from both clouds before any traffic moved. Bedrock spend of about $160K a month counts toward its AWS commitment, and the company holds the runbook, code, Terraform, guardrail test set, and evaluation tooling.
Ready to Transform Your HealthTech Business?
See how New Math Data can transform your HealthTech business with AWS-powered innovation.