Contact
Software engineer who builds AI systems
Ten years of production software: Barberan, Newshore, Voxel (an Amadeus company), Materialise, Hitachi Energy, and Wizeline for a tier-1 US bank. Industrial, travel, invoicing, medical, banking. Regulated and not. Two-person teams and large orgs.
Architecture as practice, not a title: C4 diagrams, Backstage docs and decisions taken across several teams at Wizeline; systems designed and owned from nothing at Voxel and Barberan.
The through-line is domains where being wrong has consequences: a wrong payment, a wrong medical record, a wrong instruction on a factory floor. That instinct is what I now point at AI systems. A retriever that refuses to answer rather than guess without a grounded source is the same instinct, wearing different clothes.
the evidence, at three altitudes
Each one names the alternative that lost and the price of the one that won. Including the predictions that measurement proved wrong. Those are payments-rag; oncall-triage keeps 17 more in its repo.
read the decisions →A retriever that cites the page it answered from, an alert-triage agent fed by three cloud estates, a browser extension on the store, and this site. All live, all documented down to the trade-offs.
see how they are built →read the payments-rag case study →Techniques measured and then declined on the evidence. A "no" is a finding, and it is recorded like one.
see what was tested →where the edges are
- I build systems around models. I do not train them.
- AI experience is three months: 13 articles, two systems in production, one extension on the store. The judgment it rests on is ten years.
- Large-scale distributed AI infrastructure is not something I have demonstrated.
open to B2B engagements and roles · start within a week of signing