Research & Resources
Applied research and reference architectures.
Applied Research
From Layered Controls to Action Integrity
Author: Manuel Tomas Estarlich
Applied technical working paper, August 2026
Research conducted and authored by Manuel Tomas Estarlich. The findings inform LevelUp360’s approach to applied AI security, workflow governance and verification.
The paper examines an implemented multi-agent system and the gap between controls that work locally and evidence that an authorised action remained the same through execution and verified outcome. It defines that missing property as stateful action integrity and sets out the next bounded experiment; the proposed intervention has not been tested.
Status: Author-produced working paper; not peer reviewed or independently validated.
Supporting material: Public evidence package | Architecture companions | Release manifest | GitHub repository
Reference Architectures and Control Patterns
Production-grade reference architecture for Agentic AI on Azure. Includes Microsoft Agent Framework implementation and security patterns.
View Architecture & CodeProduction-grade AI platform on the Microsoft enterprise AI stack. Designed for EU-regulated industries with document-intensive operations. Includes architecture decisions, design rationale, and implementation guidance.
View Reference ArchitectureResearch Principles
We publish the question investigated, the evidence used, negative and incomplete results, explicit limitations, and the next falsifiable question. Where source material must remain private, we state the evidence boundary rather than imply independent verification.