Back to Insights & Impact Industry Insight

Designing a Multi-Dimensional Agentic AI

for Facilities Inspection

Designing a Multi-Dimensional Agentic AI

Abstract

In Singapore’s building and infrastructure sector, the use of AI is gaining traction, especially in the inspection of façades and structural elements. Most current systems are able to detect visible defects using photos, thermal imaging, or LiDAR scans, but these AI tools still rely heavily on humans for interpretation, decision-making, and regulatory compliance. While automation has helped speed up data collection, the next step is to build AI that goes beyond surface-level analysis — to assist engineers with insights that are contextually sound, technically accurate, and aligned with local standards. This paper explores how multi-dimensional agentic AI can support that transition and proposes a practical path to develop such systems for real-world use in Singapore.

Introduction

Over the past few years, we've seen the steady adoption of drones, sensors, and AI in the inspection of building façades and critical structures. These tools are now part of the workflow for periodic inspections, offering better access to hard-to-reach areas, faster turnaround times, and more consistent data capture. However, while technology has helped reduce manual workload, the final analysis and engineering judgment still fall squarely on the shoulders of qualified professionals.

Current AI systems are good at detecting cracks, spalling, corrosion and other visible signs of wear. But beyond simply flagging defects, they lack the deeper understanding needed to assess risk, prioritise action, or recommend practical remedies. What’s missing is a system that can think more like an engineer — considering causes, long-term implications, and code-based standards — not just pointing out what’s wrong, but helping to decide what to do next.

Understanding the Limitations

Today’s AI models function mainly as detection tools. They highlight issues based on image data, but they often fail to link those issues back to the building's design, history, or usage patterns. For example, a crack might be marked out as a defect, but there’s little insight into whether it’s superficial or structural, recent or longstanding, isolated or part of a broader failure. In many cases, engineers end up treating AI results as preliminary findings that need to be double-checked — instead of reliable support that can inform professional decisions.

The current systems also struggle when multiple data types are involved. A façade inspection might include visual data, thermal readings, and laser scans. Each of these data streams tells part of the story, but AI tools typically don’t combine them in a meaningful way. Without this integration, the analysis remains shallow and fragmented.

The Engineer’s Role and Why It Still Matters

In Singapore, periodic inspections are not just routine tasks — they are formal processes with clear legal, technical, and safety responsibilities. Professional Engineers and Certified Inspectors follow strict guidelines, conduct hands-on testing, and apply their experience to interpret what they see. They look beyond surface defects and consider the building’s age, materials, load paths, environment, and maintenance history. They use tools like rebound hammers and ultrasonic testers to check what’s not visible and refer to structural codes and past cases to make sound recommendations.

These are skills and judgments that AI does not yet possess. But with the right design, we can start building AI systems that learn from these expert practices and gradually support them in more meaningful ways.

Moving Towards Multi-Dimensional, Agentic AI

The next generation of AI for inspections needs to be more than just a detector. It should be able to understand relationships between defects, identify underlying causes, and foresee potential failures. To do this, the AI must process different types of input data — not only images, but also sensor readings, material properties, and environmental conditions.

More importantly, it must also be trained to reason. For instance, when a pattern of spalling appears across a building’s external wall, the AI should not only detect each spall, but also assess whether the pattern points to deeper water ingress or reinforcement corrosion, and suggest possible next steps based on past inspection data and known deterioration models.

This kind of AI needs to be built on a framework of engineering knowledge — including design codes, inspection standards, and practical repair methods. It must also learn from feedback loops. When engineers adjust the AI’s recommendations, the system should take note and refine its logic. Over time, it will evolve to provide insights that are increasingly aligned with how professionals work on the ground.

Human-AI Collaboration, Not Replacement

At no point should AI replace engineers or take over regulatory roles. Instead, it should be built as a system that supports professionals — offering a second pair of eyes, helping to process large volumes of data, and highlighting patterns that might otherwise be missed. The goal is not full autonomy but guided assistance. The AI should explain its reasoning, present alternative views, and leave room for human judgment.

This approach will allow engineers to focus on complex decision-making while relying on AI to handle routine detection, risk modelling, and comparison with historical cases. It creates a partnership where both sides — human and machine — bring their strengths to the table.

A Practical Path Forward

To realise this vision, the development must be grounded in real-world inspection workflows. Pilot projects should involve engineers and inspectors from the beginning, so the AI learns from actual practice, not textbook theory. The system should also integrate well with platforms already used in Singapore, such as BIM models, digital twins, and SAC TR78-compliant reporting systems.

As part of the development, we must also ensure transparency. Any AI system used in an inspection must clearly show how it arrived at its conclusions, what data it used, and where its limitations lie. Only then can it earn the trust of industry professionals and regulators alike.

Conclusion

Singapore is well-positioned to lead in this area. With a strong regulatory framework, experienced engineering community, and increasing demand for smart building management, the stage is set for multi-dimensional AI to play a larger role in inspections. But the key is to design these systems not just as tools, but as partners — systems that can grow to understand our local conditions, standards, and professional ways of thinking.

By developing AI that can interpret, reason, and recommend — not just detect — we can create a future where inspections are not only faster and safer, but also more insightful and reliable. In doing so, we also free up engineers to focus on what matters most: applying their expertise to protect the built environment and ensure public safety.

About the Author

Eugene Choy is the Founder and Managing Director of NovaPeak Pte Ltd, creators of LiveInspect.AI, an AI-powered platform for drone-based periodic façade inspections and infrastructure monitoring.

Contact us to learn more about our solutions or to collaborate on future initiatives.

Get In Touch