LLMOps.Pro · ABOUT · IA, not AI
I spent 25 years making automated systems
an auditor would trust. Now I build AI the same way.
I'm Mauro Moro — an IT/OT computerized-system validation (CSV) and MES specialist in regulated pharma manufacturing. LLMOps.Pro is what happens when someone whose job is proving systems are trustworthy turns to AI agents.
IA, not AI
The human stays in the loop — and on the hook.
The industry is selling AI that replaces the expert. On the regulated plant floor I've watched what actually happens when you put an unaccountable system into a process that matters: it doesn't get adopted, because nobody can stand behind it. You can't tell an inspector “the model decided.”
So I don't build AI to replace the expert — I build it to augment one. The idea isn't new: sixty years ago Doug Engelbart called it augmenting human intellect — tools that amplify human capability instead of substituting for it. Intelligence augmentation, not artificial intelligence. The human keeps the judgment and the accountability; the agent does the toil — drafting, retrieving, cross-checking — and shows its work, every time.
I call the result the Augmented Human: a domain expert moving at machine speed, with a cryptographic receipt for every step.
One idea, found through three projects.
LLMOps.Pro didn't start as a plan. It started as the same instinct, applied wherever I could reach.
curriculum.agent · the first
A grounded-RAG agent over a single private knowledge base — my own career. It answers only from the record, with citations. Proof to myself that an agent could be useful and accountable on private data. (It's the chatbot below.)
ComplianceGxP · the serious one
Then I took it to the work I know best: an AI co-pilot for GxP compliance authoring — protocols, deviations, CAPAs, regulatory assessments — every output grounded and anchored to a tamper-evident ledger. The discipline I'd applied to validating MES, now applied to the AI itself.
The open lab · in public
A parallel, sovereign experiment: an agent that runs a small business in public, holds its own wallet, pays for its own tools, and logs every action where anyone can inspect it. The governance I practise under NDA, demonstrated in the open.
Different domains, one spine: agents you can ground, govern, and audit. That spine is the company.
What I believe
Validation discipline, applied to AI.
This is the part most AI builders skip and I can't. On the regulated plant floor you don't earn trust by being impressive — you earn it by being provable. The same rules carry over:
Grounded. No answer without a source. The agent reasons over your data, not the open web, and cites where every claim came from.
Governed. A written contract for what the agent must, must not, and can do — enforced on every run. (This became AgentContract, which I open-sourced.)
Auditable. Every decision anchored to a sequential SHA-256 ledger — tamper-evident, append-only, reconstructable for an inspector.
Human on the hook. The agent proposes and prepares; the expert decides and signs. Augmentation, not autopilot.
Show the work. If it failed, it says so, with the evidence. “Done” means verified.
GAMP 5 · 21 CFR Part 11 · EU GMP Annex 11 · ICH Q7 / Q10 — not constraints I tolerate, the shape of how I already think.
The record
45 years, on paper.
The technical résumé for the quick read, or the long-form story for the rest.
curriculum.agent
Ask my experience anything.
The first agent I built still runs. Ask it about my pharma validation work, SCADA/PLC platforms, MES, or software background — it answers from the record, with citations. It's the same grounded-RAG capability behind Pillar 3 of the live platform, pointed at my career instead of GxP frameworks — the easiest way to watch grounded AI work on a real private knowledge base. The same thing I'd build on yours.
Get in touch
Talk to us.
For a pilot, a partnership, a press question, or just hello.