Code is the easy part. The hard part is the
that ship it, trust it, and outlive their creators.
Three beliefs. Twenty-one years of proof.
Reads a TypeScript codebase and reports what’s actually there — hotspots, blast radius, bus factor, duplication — from AST and git history, not estimation. No model in the loop. No code leaves the machine. Every number is verifiable against the repo you point it at.
I built it because the things I care about inside a company — whether the team can operate without the original engineer — are the same things I wanted to measure outside of one.
A natural-language interface to Azure Data Explorer where the LLM is never allowed near the parts that need to be trusted. Clusters, schemas, auth, and metric definitions live in declarative YAML config; the LLM only writes filter clauses. Zero hallucination isn’t a claim — it’s an architectural property of the design.
I built it because the question I kept hearing inside a company — can we trust this LLM output? — has a better architectural answer than I was hearing in public. This is the shape of that answer.
Proved by Cross-service tracing framework · RAI Champion · Pulse
Proved by Roche cobas · 1,400 tubes per hour · clinical diagnostic pipeline
Proved by Wipro automation layer · Nokia protocol suites · Stuttgart routing framework
AI is becoming the heart of everything. I drive the layer beneath it — the one that makes a wrong autonomous decision traceable, auditable, and structurally non-repeatable. A failure two services deep is now a link in a trace, not a three-hour forensics exercise.
1,400 tubes an hour. I architected the validation framework where a misidentified STAT sample — a cardiac troponin joining a routine queue — became mathematically impossible. When the digital thread broke at Siemens, the wrong thing got built downstream. I made both categories of failure structurally unreachable.
Both sides of the door. Fremantle — fully exposed, hackable, cellular stacks visible. Windows Phone — locked, platform-owned. I built test suites on both that made platform-level failures structurally impossible, regardless of what the platform allowed me to see.
Trust is architectural
Speed is useless if the direction is untrusted
Correctness has a face
Real-world impact is complex and requires absolute visibility
Make it impossible
The only acceptable outcome is the end of the problem
The teams were fast, but they were running in circles. We were fixing the same regression every three weeks. That was the year I stopped valuing the fix and started valuing the extinction of the error.
Protocol suites don’t lie. If the trust isn’t in the handshake, the packet doesn’t move. I realised then: you cannot decorate a system with trust later. It is either in the architecture, or it isn’t there.
I stood in a lab in Stuttgart and watched the tubes move. 1,400 lives an hour, distilled into telemetry. Correctness stopped being a unit test; it became a face.
AI is becoming the heart of everything. The work is no longer about building the agent. It’s about building what makes the agent trustworthy enough to be. Twenty-one years of making things impossible, giving correctness a face, and treating trust as architecture — this is where those three beliefs stop being personal convictions and become the job.
Unedited · from peers
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.
Technical and business acumen always at the high end — a role model for his peers.
A true professional and a true gentleman. Would work with him again without hesitation.
Actively driving test automation across both protocol-level messaging and the user interface.
Versatility and out-of-the-box thinking — an exceptional knack for identifying critical defects early.