Date: |
15 Jun 2026, 5.30PM – 6.30PM |
duration: |
1 hr |
Location: |
Online |
Cost: |
Free event |
The session provides a practical, evidence-based perspective on evaluating AI as a reliable tool rather than a replacement for established safety processes.
In safety-critical environments, a system that "guesses" isn't an innovation - it’s a liability. Because Large Language Models (LLMs) operate on static statistical patterns, treating them as standalone solutions introduces unacceptable risks like data hallucinations and rigid knowledge cutoffs. But what happens if we look past the generic AI hype and treat these models as components within robust, deterministic engineering frameworks?
Join the New Zealand Society for Engineering Safety (NZSES) for an insightful presentation with Paul Leroy, a veteran IT consultant and Google Cloud Trainer. Drawing on his foundational background in industrial electrical engineering, over 15 years of experience architecting cloud systems, and his role as the Policy Working Group Lead for the Thames Valley AI Hub, Paul bridges the gap between heavy industry, governance, and modern data science.
Rather than presenting AI as a foregone conclusion that will magically solve all operational challenges, this session explores how advanced AI pipelines might serve as viable, structured avenues to enhance risk management and compliance.
Key areas for discussion include:
- Retrieval-Augmented Generation (RAG): How anchoring LLMs directly to your organization’s verified manuals, schematics, and regulatory text can act as a safeguard against inaccurate AI outputs.
- Document Synthesis: Evaluating the potential of AI pipelines to assist in reviewing complex safety procedures and highlighting potential compliance gaps.
- Computer Vision and Multimodal Tools: Exploring the frontier of using models like Gemini to analyze site photographs, evaluating their current viability for identifying visual safety and PPE violations.
Whether you are looking to cautiously audit automated compliance workflows or simply want to understand the realistic boundaries of digital safety systems, this session offers a practical, engineered roadmap for evaluating AI as a predictable asset for operational excellence.
This presentation challenges the idea of using Large Language Models (LLMs) as standalone solutions in safety-critical environments, highlighting risks such as hallucinations and outdated knowledge. Paul Leroy will explore how AI can be integrated into structured engineering frameworks through approaches like Retrieval-Augmented Generation (RAG), document synthesis, and computer vision to support risk management, compliance, and operational safety.
The session provides a practical, evidence-based perspective on evaluating AI as a reliable tool rather than a replacement for established safety processes.
In safety-critical environments, a system that "guesses" isn't an innovation - it’s a liability. Because Large Language Models (LLMs) operate on static statistical patterns, treating them as standalone solutions introduces unacceptable risks like data hallucinations and rigid knowledge cutoffs. But what happens if we look past the generic AI hype and treat these models as components within robust, deterministic engineering frameworks?
Join the New Zealand Society for Engineering Safety (NZSES) for an insightful presentation with Paul Leroy, a veteran IT consultant and Google Cloud Trainer. Drawing on his foundational background in industrial electrical engineering, over 15 years of experience architecting cloud systems, and his role as the Policy Working Group Lead for the Thames Valley AI Hub, Paul bridges the gap between heavy industry, governance, and modern data science.
Rather than presenting AI as a foregone conclusion that will magically solve all operational challenges, this session explores how advanced AI pipelines might serve as viable, structured avenues to enhance risk management and compliance.
Key areas for discussion include:
- Retrieval-Augmented Generation (RAG): How anchoring LLMs directly to your organization’s verified manuals, schematics, and regulatory text can act as a safeguard against inaccurate AI outputs.
- Document Synthesis: Evaluating the potential of AI pipelines to assist in reviewing complex safety procedures and highlighting potential compliance gaps.
- Computer Vision and Multimodal Tools: Exploring the frontier of using models like Gemini to analyze site photographs, evaluating their current viability for identifying visual safety and PPE violations.
Whether you are looking to cautiously audit automated compliance workflows or simply want to understand the realistic boundaries of digital safety systems, this session offers a practical, engineered roadmap for evaluating AI as a predictable asset for operational excellence.
