Tactics are only worth the strategy they serve.
Two decades read like a distributed trace: four spans, three countries, and a return to where it started.
Wipro automation, then Nokia's N900 and N9. The same regression kept coming back, so I stopped valuing the fix and started designing out the error.
RCSe validation on SIP signalling. If the trust isn't in the handshake, the packet doesn't move.
Live clinical lines running ~1,400 samples an hour. A failing test stopped being a unit test and became a patient.
Back where it started.
Nine years from senior engineer to engineering lead: tracing that became the org standard, then the agent harness behind an enterprise AI copilot.
Trust is architectural.
You can't add it later. It is in the design, or it isn't there.
Proof · Agent harness, governed by design ↓Correctness has a face.
In a clinical lab, a failing test is a patient. I've never stopped seeing it that way.
Proof · Roche, 1,400 samples an hour ↓Prevent the category, not the instance.
If the same bug comes back, the design is wrong, not the code.
Proof · Tracing that became the standard ↓Every engineering decision serves a strategy. Here is the tactical call, and what it changed.
Correlation IDs across a front-end and back-end boundary that used to swallow tickets.
Support SLA cut · adopted as the org-wide tracing standard
Set platform and AI direction for a 15-engineer team. Versioned tool contracts mean the model invokes tools and never writes unvalidated state, built on Semantic Kernel, APIM guardrails, Azure AI Foundry and MCP.
TargetShorter release test cycles · 75% less developer toil
Led integration of the cobas sub-components (control unit, sample supply, module buffers) and the Python validation framework behind them.
Samples an hour · releases held on live clinical lines across four product families
RCSe validation at protocol level: SIP signalling against GSMA RCS specifications.
Carrier certifications · Orange, Deutsche Telekom, Vodafone
Contributor · Microsoft Agent Framework
Plain-English questions to Azure Data Explorer. The model writes only the filter clause; clusters, schemas, auth and metric definitions live in config it can't touch.
Reads a TypeScript codebase and reports what's actually there. No model in the loop, no code leaves the machine, every number checks against the repo.
Data structures, algorithms and system design, worked through the way real AI systems use them: agent DAGs, KV caches, batching, RAG. Every problem comes with sources and tested C# code.
GitHub ↗“A thoughtful collaborator who uplifts those around him. His combination of technical depth, adaptability, and natural leadership make him an incredible asset to any team.”
I build teams that deliver value to the business, not just software. Looking for the next one.