Pre-Sales to Production · AI Product Lead & Applied AI Builder
For seven years my work has started in the same place: in a room with an enterprise that hasn't decided yet. Pre-sales and discovery, then scoping what is actually buildable, then owning whether it holds up in production. Four consultancies, twenty industries, the same motion each time — now applied to AI.
What is unusual is not any one of those. It is that they sit in one person: I can hold the conversation with the executive deciding whether to trust AI at all, build the thing that proves it, and then measure whether it is actually correct rather than merely convincing. Most rooms have someone who can do one of the three.
I don't hand a concept to an architect and wait for a build. I design and build the working reference system myself, using coding agents as my engineering layer — the integrations, the evaluators, the tests, the security review, the deployment.
So engineering never receives a specification. It receives something already running and already measured, and adapts a proven implementation instead of starting from a document. That is the whole shift: the risky, ambiguous part is finished before the team is asked to commit to it.
Agent-generated does not mean ungoverned. Nothing counts as working until it survives checks that are code rather than judgement: deterministic scripts that pass or fail, secret scanning, and a security review run as an adversarial panel — independent agents from two different model vendors search for findings, then a separate panel tries to refute each one, and only findings that survive a quorum are treated as real. Live access controls get verified against the running system, not just the source, because the most serious thing I have found this way was in configuration while the code above it was clean.
Independent at 0→1. Collaborative at production scale.
Agent networks that handle the full cycle — from requirements to deployment — with human-in-command control and full decision traceability.
In the room before the decision is made: discovery with an enterprise that has not scoped its own problem yet, then the technical win, then embedded delivery — building 0→1 and owning whether it survives production.
Red-teaming, resistance testing, RAG quality, and root-cause attribution — proving AI is correct, not just plausible.
Platforms that explain why decisions were made — built for regulated environments, audit requirements, and long-term maintainability.
RAG pipelines, LLM tool-calling, MCP integrations, policy gates, audit logs, and role-based access at enterprise scale.
Product, BA, and delivery functions built from zero to scale — across EU, US, and LATAM teams in 20+ industry domains.
The AI systems on this site are recent and most of them are 0→1. They sit on ten years of the slower kind of product work — the kind measured in quarters and in other people's careers.
0→1 is what I am known for now. It is not the only speed I have run at.
Before focusing on AI platforms, I led the design and delivery of complex enterprise systems across regulated and data-intensive domains:
I focus on operational AI platforms that are fast to build, safe to run, and trusted by the business.
Roles where AI becomes core enterprise infrastructure — not a feature, but a system teams rely on every day.
Let's build something that matters.
Get In Touch