Presentation: Decision Models in Agentic Architectures: From Production to Agent Skills
| Source: InfoQ AI/ML
Tags: DMN, agentic-AI, NeMo-Guardrails, enterprise-AI, Drools, explainability, QCon
QCon AI presentation from Aletyx CEO Alex Porcelli demonstrates how wrapping LLM agents with DMN (Decision Model and Notation) rules creates auditable, deterministic decision paths — letting business teams own governance logic while engineers maintain architectural control.
Details
Alex Porcelli, co-founder of Aletyx and a 15-year contributor to Drools, jBPM, and Kogito, presented at QCon AI on combining deterministic DMN decision models with LLM-based agents to solve enterprise AI's accountability gap.\n\nThe core argument: non-deterministic LLM outputs are unacceptable for high-stakes decisions like loan approvals, compliance checks, or clinical triage. DMN provides a formal, auditable layer where business analysts can express rules in a standard notation, independent of the model underneath. LLMs handle the unstructured reasoning; DMN handles the guardrails and the explainability record.\n\nPorcelli also integrates NVIDIA NeMo Guardrails into the architecture for real-time output filtering, creating a three-layer stack: LLM reasoning → DMN governance → NeMo safety rails. The talk is grounded in production experience rather than theory, drawing on enterprise automation deployments.\n\nThe presentation is from QCon AI, a practitioner-focused conference. Content is available as a 43-minute video with slides. The source is InfoQ, which has a strong track record for engineering-quality conference coverage.