How Dexcom Built an AI-Enabled Product with Ketryx
The first company to bring generative AI to glucose biosensing maintains continuous compliance and traceability with Ketryx.
Dexcom is the global leader in glucose biosensing, serving millions of people worldwide. Its over-the-counter biosensor, Stelo, leverages generative AI to turn real-time glucose data into personalized insights on how sleep, meals, and activity shape a person’s health. With Ketryx, the Stelo team achieved live end-to-end traceability, item-level review and approval, and automated design control documentation, all within the tools they were already using (Jira, GitHub, TestRail, Jama).
Impact
Pain Points: Manual Compliance Bottlenecks
As a leader in AI-enabled devices, Dexcom has an ambitious product roadmap that generates traceability and change faster than manual compliance can absorb. While building Stelo 3.0 under a tightening regulatory landscape, the team identified four compliance hurdles to achieve their roadmap goals:
- Traceability Lagged Behind Development: Maintaining end-to-end traceability across disconnected tools meant quality and compliance always chased development, while reconstructing a single trace path required four to six documents assembled by hand.
- Document-Gated Control Points: Moving from development complete into formal testing took as long as three weeks because waterfall processes locked design inputs, outputs, test plans, and approvals at the document level.
- Compliance Separated from Engineering Work: Developers collected evidence in separate systems from their preferred tools. Previous attempts at building homegrown compliance solutions had required significant validation and development resources.
- Manual Document Assembly: Release and 510(k) submission documents were generated slowly by hand, delaying product cycles.
Solution: Control Layer Automating Traceability and Enabling Agile Development
Dexcom deployed Ketryx across the Stelo team’s tools, with the traceability, controls, and human-in-the-loop rigor that AI-enabled products demand:
- Automated End-to-End Traceability: Ketryx integrates Jama, Jira, TestRail, and GitHub into a continuously updated traceability matrix, with automated checks that confirm the correct test case versions map to requirements.
- Item-Level Review and Approval: Control moves from the document level to individual items, allowing the Stelo team to review and approve incrementally throughout the release cycle. Ketryx automatically assesses the change impact of each item and flags downstream approvals across every connected system.
- Tool-Based Compliance: Ketryx is validated out of the box and captures evidence automatically from the tools teams already use, making compliance a byproduct of engineering work rather than a separate step at the end.
- Compliance Artifacts on Demand: Because Ketryx stores all auditable compliance evidence, teams can automatically generate trace matrices and submission artifacts, including requirements and risk management documentation.
Business Outcomes: Continuous Compliance
With additional controls and automated traceability and documentation, the team saw compressed cycle times and improved product quality and safety:
- Improved Cycle Times: Automation of test traceability documentation has significantly reduced manual effort and cycle times.
- End-to-End Trace Coverage: Traceability across requirements, test cases, and code is maintained and checked as work happens, so the team continuously knows where coverage stands.
- Agile Releases: Item-level control enables the team to release incrementally by approving items as they are ready and re-approving only those impacted by changes.
- Cross-Team Shift Left: Engineering and quality both work from a continuous and consolidated knowledge graph, allowing teams to identify coverage gaps earlier and reduce rework overhead.
“Ketryx enables asynchronous development of our product and the documentation required to stay compliant, move fast, and make high-quality AI products at scale.”

“Ketryx was the only solution that met our needs, because it gave us a deterministic AI control layer that interweaves compliance into development. If you’re a life sciences company building with AI, this is how you do it safely.”

.png)
