Agentic AI and Multimodal Expert Systems for Intelligent Infrastructure Decision Support: The CIAS Architecture

This package accompanies a PRISMA-informed systematic mapping review that introduces the Integrated Civil Infrastructure Autonomy Stack (CIAS)—an evidence-derived expert-system architecture integrating multimodal perception, vision–language models, knowledge-driven Digital Twins, and agentic AI for intelligent infrastructure decision support. The associated manuscript is prepared for Expert Systems with Applications. Manuscript title: Agentic AI and Multimodal Expert Systems for Intelligent Infrastructure Decision Support: The CIAS ArchitectureVersion: 4.1.0-ESWALicense: CC BY 4.0 Evidence summary: six Scopus streams identified 434 records; 423 unique records were screened (328 Include / 45 Maybe / 50 Exclude). A prioritised full-text synthesis corpus of 110 papers (Tier-1 = 43; Tier-2 = 67) underpins the coded analyses; 218 Include records remain in the breadth map only. Headline multi-label counts in the coded corpus: Digital Twin 71; Multimodal AI 58; LLM 37; VLM 19; Agentic AI 13 (CIAS Layer 3 = LLM and/or VLM union = 37). Package contents: 01 — Master Evidence Database: full coded dataset (CSV + XLSX) for the 110-paper prioritised full-text synthesis corpus, including Scopus metadata, screening fields, and paradigm/asset/sensing coding variables. 02 — Search Strategy: six thematic Scopus TITLE-ABS-KEY Boolean strings (Agentic AI; Multimodal AI; Agentic + Multimodal; Autonomous Inspection; Infrastructure Asset Management; Digital Twin). 03 — Screening Files: PRISMA screening ledgers and RIS exports (Included / Maybe / Excluded), duplicate-removal records, and title/abstract screening summaries. 04 — Coding Documentation: coding dictionary for all analytical variables, including the operational definition of Agentic_AI used throughout the synthesis. 05 — Quality Assessment: Stage A triage and Stage B reporting-completeness appraisal scores for the full-text coded corpus, plus the exact Stage B 0–2 × 5 scoring rubric. 06 — PRISMA: PRISMA 2020 flow diagram (figure + final count summary) documenting identification through inclusion for the two-stage mapping design. 07 — Methodology Guide: evidence-extraction and coding protocol used to construct the database (Markdown + PDF). 08 — IRR Worksheet: inter-rater reliability dual-coding worksheet (30-record sample) and κ-computation script. Coder A is complete; Coder B is currently incomplete and is disclosed as such—reported prevalence figures should be read as single-coder estimates pending independent verification. Cohen’s κ is not claimed. 09 — CIAS Architecture: architecture figures (Twin-before-Agent expert-system stack, agentic inspection loop, graphical abstract, review workflow) and Table 11 comparing CIAS with Sense–Plan–Act, OODA, MAPE-K, autonomous-driving stacks, robot sense–act patterns, and Digital Twin lifecycle models. 10 — Supplementary Coded Bibliography: full bibliographic listing (DOCX + XLSX) of all 110 papers in the prioritised full-text corpus, with paradigm flags (Agentic AI, Multimodal AI, LLM, VLM, Digital Twin).