Building a Governed AI Foundation for Clinical Genomics

How TeraSky built the Microsoft Azure foundation for Geneyx’s AI-assisted variant interpretation, with privacy, traceability, and human review built in.

Geneyx helps genetics labs, hospitals, and life science companies turn genetic sequencing data into clinical insights.

A key part of that work is variant interpretation. Geneticists review evidence about a genetic variant, compare it with trusted medical sources and published research, and write the interpretation that may ultimately appear in a clinical report.

Geneyx saw a clear role for AI in helping draft and summarize that work. But introducing AI into a regulated clinical product required a foundation that could support privacy, traceability, testing, and human oversight from the beginning.

TeraSky worked with Geneyx to build the governed Microsoft Azure foundation needed to bring its clinical AI roadmap into the product, starting with variant interpretation.

The Challenge

Variant interpretation is detailed, evidence-heavy work. Geneticists review sources such as ClinVar, OMIM, and published literature, then turn that evidence into a clear clinical explanation. The process takes time, and the final wording can vary between analysts.

AI could help accelerate the drafting process, but Geneyx needed tight control over its introduction. Clinical data had to remain inside the Geneyx environment, every AI interaction needed to be traceable, and analysts had to remain responsible for final approval.

There was also a broader platform question. Geneyx had several AI use cases on its roadmap, so building each one independently would create a new set of security, governance, and integration decisions every time.

That made the AI foundation itself a core part of the project.

The Architecture

TeraSky designed a dedicated AI landing zone within Geneyx’s Microsoft Azure environment, using Microsoft Foundry as the central layer to host, govern, and monitor AI workloads.

The environment was designed around Geneyx’s clinical requirements. AI services are kept private, access is controlled through Microsoft Entra ID, and managed identities eliminate the need to store credentials between services.

TeraSky translated those requirements into the underlying Azure architecture needed to support Geneyx’s clinical AI workflows. The platform provides the secure compute, identity, connectivity, retrieval, and observability layers on which those workflows run. Geneyx retained ownership of the clinical layer itself, including its instruction set, guardrails, evidence-trace and provenance design, and usage metering.

Requests enter through a controlled application layer, where the system can retrieve relevant information from Geneyx’s knowledge base and use Azure OpenAI to generate draft clinical text. Rules that need to be fixed, predictable behavior remain outside the model itself, so the AI is used for reasoning and drafting, while explicit logic handles requirements that should always behave the same way.

The architecture also provides the infrastructure needed for tracing and auditability across AI interactions. Infrastructure is managed as code, giving Geneyx a controlled and repeatable Azure foundation while the deployment pipeline for this service is finalized.

With that Microsoft foundation in place, Geneyx could bring its first clinical AI workflow onto the platform.

Technologies Used

Microsoft Azure丨Microsoft Foundry丨Azure OpenAI丨Microsoft Entra ID丨Azure AI Search丨Terraform丨Git

The First Use Case

The first use case is a Variant Interpretation Agent.

Geneyx sends variant details and analyst-selected evidence into the platform. The system retrieves supporting information from Geneyx’s knowledge base, and Azure OpenAI prepares a draft interpretation using a controlled clinical template.

Before the draft reaches the analyst, the system checks that required references such as ClinVar, OMIM, and PMID are present and that the response follows the expected structure. The analyst then reviews, edits, and approves the interpretation.

Geneyx’s AI team maintains a regression testing framework built around a golden-set corpus, replay harness, scoring methodology, and benchmark records. Prompt and model changes can be tested against known-good outputs before release, helping the team evolve the clinical AI layer on top of the TeraSky-built foundation while verifying expected behavior over time.

Validation is performed using de-identified clinical data.

The Impact

The pilot provides Geneyx with a structured foundation for introducing AI into its clinical platform.

The Variant Interpretation Agent demonstrates what that foundation makes possible. Geneyx can run a real clinical AI workflow within a private Azure environment with controlled identity, retrieval, connectivity, and observability, while the geneticist remains responsible for the final interpretation.

The larger value is the reusable Azure platform underneath it. TeraSky provided Geneyx with a foundation that supports additional AI use cases without having to rebuild the core infrastructure, security, identity, and observability layers each time. Geneyx can continue to evolve its clinical intelligence and controls on top of that common platform.

For Geneyx, that creates a more practical way to expand AI across the product while keeping clinical and regulatory requirements visible throughout the development process.

“AI has been on our roadmap for some time, but we needed to know exactly where the data sits, what happens on every call, and how the analyst stays in control. TeraSky provided the private Azure foundation needed to support that level of control and to bring our first clinical AI use case onto the platform. It gives us a foundation we can extend as we add new AI capabilities.” —Yaron Mazor, CTO, Geneyx