H3 Zoom develops AI-powered inspection and asset-intelligence solutions for the built environment. We help teams detect, organise, and report defects more consistently, while creating structured digital records for quality assurance, compliance, and lifecycle asset management.
When we joined BEAMP, we were looking for an industry partner who could provide the domain expertise, real site access, and operational rigour needed to test our technology under genuine construction conditions. In City Developments Limited (CDL), we found exactly that.
High Standards, High Volume
CDL is a leading global real estate company with a network spanning over 160 locations in nearly 30 countries and regions. Their residential projects are built to high quality standards through rigorous QA/QC inspections.
After the Building and Construction Authority (BCA)'s Quality Mark assessment, CDL's Quality Inspection and Customer Service teams conduct final aesthetic and functional checks before units are handed over to homeowners.
These inspections are thorough, but they are also manual, time-consuming, and documentation-heavy. Inspectors walk through each unit, identify defects, mark them on site, and then input records into CDL's defect management system.
With each development comprising dozens of units, the administrative burden adds up quickly. CDL wanted to explore whether technology could automate defect detection and reporting — reducing manual effort and inspection time without compromising quality.
One Workflow, from Capture to Report
Over a 14-month development period, we worked closely with CDL's QA team across two live residential developments — Tembusu Grand and Lumina Grand — to develop and refine an AI-assisted inspection workflow.
What makes our solution different is that it combines lightweight mobile capture, AI-assisted defect detection, building-element classification, spatial mapping, and automated reporting in one end-to-end workflow, while keeping the inspector in control of final verification.
The process works like this: an inspector walks through the unit with a mobile device, capturing images at regular intervals. Our AI engine analyses the images to generate candidate defect observations, classifies each one by building elements — wall, ceiling, floor, cabinet, window — and maps them spatially to the unit's floor plan.
The platform then compiles everything into a structured inspection report within 48 hours, complete with defect images, locations, and classifications.
We also developed an immersive navigation interface — similar to Google Street View — that lets users virtually walk through the inspected unit and review defects in their surrounding context, rather than examining isolated photographs.
Testing Against the Real Benchmark
Throughout the project, we worked closely with CDL's QA team to establish a shared understanding of what counts as a valid defect and how AI-generated observations should be compared against CDL's manual inspection records.
Multiple review sessions helped define the evaluation rules before any benchmarking began — ensuring that performance metrics reflected meaningful inspection outcomes, not differences in interpretation.
The pilot delivered strong results: approximately 77% defect detection accuracy (exceeding the 70% target), automated report generation within 48 hours, and a structured repository of around 20,000 labelled inspection images across the two developments. The platform also identified additional valid defects beyond CDL's original dataset, indicating broader defect coverage.
A key takeaway was that the original inspection-time target could not be met without compromising image quality and coverage — achieving the level of detection accuracy demonstrated in the pilot requires sufficiently complete and detailed imagery, and that process is currently constrained by capture hardware.
What’s Next
As hardware technologies continue to evolve, CDL expects the solution to become more user-friendly and scalable, making it more practical for wider inspection use and progressively reducing manual effort while improving overall inspection efficiency.
On our end, we plan to strengthen guided capture, add voice and text inputs, improve integration with enterprise defect-management systems, and scale the workflow across more residential units. Our goal is to make inspection documentation faster and more consistent — without removing the professional judgement and accountability that quality assurance depends on.
The BEAMP Difference
Participating in BEAMP gave us real-site access and CDL's QA expertise, allowing us to validate our technology, refine the workflow, and define a credible commercial pathway. Over 14 months, we were able to iterate through four different hardware configurations, test under varied site conditions, and arrive at a deployment model grounded in evidence rather than assumptions.
For CDL, the programme's strength lies in its ability to bridge industry challenges with innovative startups and SMEs — enabling solutions to be tested and validated in real-world environments before wider deployment.
Shaun Koo
CEO & Founder, H3 Zoom
