Authors

Ina WONG

is a distinguished leader in construction innovation and digital transformation, with extensive professional experience in delivering intelligent infrastructure and advancing technology adoption across Hong Kong’s public works sector. She possesses exceptional expertise and notable achievements in driving AI- and IoT-enabled solutions for major infrastructure projects, contributing significantly to the digitalization and modernization of the construction industry. In her role leading innovation and digital initiatives, Ina has overseen the entire lifecycle of technology deployment—from research and development to field implementation—establishing agile frameworks and innovation pipelines that transform conceptual ideas into deployable AI products. She successfully founded and scaled the company’s first integrated technology division, bringing together professionals from engineering, IT, and data science to deliver award-winning digital solutions. With an academic foundation in Political Science, Ina integrates strategic understanding of governance, public policy, and stakeholder dynamics into her approach to technological innovation. She has extensive experience fostering collaboration among government departments, consultants, and global innovators, ensuring that emerging technologies align with public policy objectives and national smart infrastructure goals. Leveraging her proven track record in innovation management, cross-sector engagement, and strategic policy alignment, Ina combines visionary leadership with practical execution capability—providing a powerful driving force for the digital transformation and intelligent development of Hong Kong’s construction and infrastructure sectors.

CHENG Ho Lam Holly

Principal Geologist, C M Wong & Associates Ltd.

CE32/2024(GE) Identification of Vulnerable Hillside Catchments Affecting Existing Important Facilities on Hong Kong Island and Outlying Islands – Feasibility Study
Development of Upgrading Works Cost and Time Estimator for Landslip Prevention and Mitigation Programme (LPMitP)
Establishment and Pilot Run of Smart Site Safety System for LPMitP
Inspection of Remaining Dwellings under Non-Development Clearance(NDC) Programme

AI-Enabled Slope Upgrading Works: From Static Records to Traceable Geotechnical Intelligence

I.H.H. Chan & K.S.T. Kwong

Landslip Preventive Measures Division 1

Geotechnical Engineering Office, Civil Engineering and Development Department, Hong Kong

B.J. Jiang, T.M. Ma, C.S. Sun & I.W. Wong

BuildWave Innovation Ltd, Hong Kong

C.T.L. Lee & H.H.L. Cheng

C.M. Wong & Associates Ltd, Hong Kong

ABSTRACT

The planning of slope upgrading works under Hong Kong's Landslip Prevention and Mitigation Programme requires reliable early-stage cost and construction programme estimates, together with comprehensive review of historical district files. In practice, this process suffers from two major challenges: structured databases do not fully capture the site-specific and historical factors affecting cost and duration, while supporting evidence is embedded in scanned multi-part files containing mixed correspondence, drawings, certificates and unrelated departmental memoranda. As engineers must investigate historical records for each slope, manual review is time-consuming and difficult to standardise.

This paper presents an integrated AI-enabled workflow that combines an explainable estimator with a traceable historical-file review pipeline. The estimator utilises structured slope attributes to generate bounded cost and duration references with explicit confidence indicators for contract-level planning. It provides a practical planning baseline before manual review, with cost estimation demonstrating stronger predictive capability and duration estimation more meaningfully expressed as bounded planning intervals. For archive retrieval, the vision-based page classification module achieved over 96.5% accuracy on 799 archival image samples, enabling reliable separation of text, drawing and other page types for downstream document understanding.

Building on this foundation, the proposed pipeline segments scanned district files into logical correspondence, drawing, submission and certificate units. Spatially related drawings are retrieved through coordinate reconstruction, while text-only evidence is identified through semantic search over slope features, project references, place names and geotechnical terms. The workflow can compress the engineer's detailed review set to around 10% of the original pages. The retrieval strategy is recall-oriented: candidate GEO comments, approvals, certificates and ambiguous evidence are retained for human validation before final registry output. Overall, the proposed system links predictive planning, comprehensive archive review and source-traceable slope history reconstruction, improving planning consistency, auditability and engineering decision support.

KEYWORDS: AI-enabled workflow, large language model, vision language model, agentic retrieval

Page updated

Google Sites

Report abuse