PDF artifact
The Prompt Governor
A governance layer for AI prompting that makes "vibes-based" prompts illegal.
What this shows: How I think about guardrails, accountability, and making AI workflows auditable.
Leonardo Pasqualin · Leo / leopaz
I design and govern systems where humans and AI can’t quietly ruin things.
I'm Leonardo Pasqualin (most people call me Leo), a builder-operator obsessed with making work predictable: clear pipelines, hard guardrails, strong journaling, and the ability to undo bad decisions—whether they came from a human or a model.
Projects & Artifacts
A small set of concrete things I use to keep AI, automation, and people from doing dumb, irreversible work.
System architecture
Siftwise — File Intelligence Architecture
A deterministic pipeline for ingest → analyze → propose → govern → execute.
What this shows: How I design AI systems with journaling, undo, and clear separation of concerns.
PDF artifact
Governor’s Checklist
A pre-flight checklist before you let humans or agents touch real workflows.
What this shows: My bias toward concrete checklists over hand-wavy "best practices".
PDF artifact
Automation Readiness Gate
A simple gate that decides if something deserves automation yet.
What this shows: How I prevent people from automating chaos and calling it efficiency.
PDF artifact
Prompt Failure Diagnostic Sheet
A structured way to debug bad AI outputs without superstition.
What this shows: My approach to turning AI "hallucinations" into measurable failure modes.
SOP
Governed SOP (Sanitized)
A real-world SOP redesigned to be observable, testable, and agent-friendly.
What this shows: How I translate messy operations into clear steps tools and agents can actually follow.
How I Work
My background is a mix of operations, systems design, and AI-first workflows. I care less about titles and more about whether the system can be trusted when things go sideways.
Governance first
I assume humans and models will eventually do something stupid. So every workflow gets guardrails, journaling, and clear ownership before it gets "optimized."
Deterministic where it matters
I like AI for analysis and suggestion, but routing, access, and execution stay deterministic. If we can't reason about it, we shouldn't ship it.
Tool-agnostic, guardrail-biased
Whether it's Next.js, n8n, or a command-line tool, I care that the system is observable, testable, and easy to roll back—not which logo is on the landing page.
No guru energy
No courses, no funnels, no inflated case studies. Just artifacts, systems, and real constraints.
Contact
If you're working on something where AI, automation, and operations collide—and you actually want guardrails, not hype—reach out.
Email: you@leopaz.dev
Replace with whatever address / alias you actually want public.