Vamsi Sai Kalasapudi
Alekya Seelam
Sneha Ganupa
Charles Tofferi- 1Department of Construction Management, University of North Florida, Jacksonville, FL, United States
- 2Petticoat-Schmitt Civil Contractors, Jacksonville, FL, United States
Research on artificial intelligence (AI) in construction over the past two decades has been dominated by sensor-driven applications such as safety analytics, progress monitoring, and sustainability modeling. These domains have benefited from abundant IoT, vision, and drone data, reinforcing an Industry 4.0 paradigm centered on physical automation and digital twins. However, this emphasis has left a major gap: the cognitive and administrative workloads that consume substantial professional time remain largely unaddressed. Documentation tasks such as reviewing specifications, assembling submittal logs, and preparing transmittals continue to rely on manual processes, despite their centrality to project delivery. This perspective paper examines this overlooked frontier through a targeted review of AI literature and a bibliometric scan of two decades of publications. The analysis shows that research on documentation automation accounts for only a small fraction of AI-in-construction scholarship compared with sensor-based domains. To illustrate the potential of document-native AI, the paper presents a case study of a submittals automation tool developed and tested with an industry partner. The system extracts submittal requirements from 1,000 to 1,500-page specifications, generates transmittal-ready Excel and Word files, and enables conversational querying. Field testing demonstrated substantial time savings and high user trust, while also revealing opportunities for improved responsiveness and workflow integration. The results show that lightweight, conversational, document-native AI can meaningfully reduce repetitive cognitive work while preserving human oversight. The paper concludes with a research agenda emphasizing reliability, multimodal document understanding, agentic workflow orchestration, and evaluation frameworks that quantify return on investment.
1 Introduction
Over the past decade, applications of artificial intelligence (AI) in the construction industry have been largely shaped by a set of prominent application areas (Xiao et al., 2018). Academic literature, reinforced by industry demonstrations, has showcased advances in computer vision–based safety monitoring, drone-enabled site mapping, image reconstruction, digital twin development, IoT integration, autonomous equipment, and field robotics, with these technologies delivering tangible benefits in safety, project planning, progress monitoring, and asset management (Guo et al., 2017; Xiao and Kang, 2021; Xue et al., 2021; Salem and Dragomir, 2022; Rangasamy and Yang, 2024; Bock, 2015). The prominence of these research areas stems largely from the abundance of sensor-based data collection opportunities and strong collaboration between academia and industry stakeholders to deploy and validate such field-facing solutions (Jang et al., 2021; Vasenev et al., 2014; Taneja et al., 2010). However, a growing perspective within the industry emerging in parallel with the rapid adoption of generative AI is shifting attention toward a different frontier: the automation of repetitive, low-value cognitive tasks that dominate much of construction administration (Gao et al., 2025; Aqib et al., 2025; Chen et al., 2025; Knyazeva and Larin, 2021; Jacques de Sousa et al., 2024). While current research emphasizes physical automation and sensor-driven insights, the day-to-day reality for many construction professionals involves extensive manual processing of documentation, submittals, transmittals, meeting minutes, and compliance records (Caldas and Soibelman, 2003). These activities consume substantial labor hours that could otherwise be allocated to high-value field operations requiring expert judgment (Franzia. AI Meets Human Intelligence aimhi.ai, 2024).
The recent boom in conversational generative AI has opened the door to an alternative approach, where natural language interfaces allow construction professionals to interact with sophisticated AI systems in plain English, thereby reducing the steep learning curve typically associated with complex project management software (Rane, 2023; Ghimire et al., 2024; Ghimire et al., 2023; Saka et al., 2024; Saka et al., 2023). In this paradigm, AI is not regarded as a job replacement mechanism, but as an assistant that frees site managers, superintendents, and project engineers from the burden of repetitive administrative duties (Amoah, 2025). By delegating routine documentation and review workflows to AI, human expertise can be focused on critical tasks such as monitoring site conditions during complex operations or resolving on-site problems in real time (Buildots, 2025). In this paper, the authors advance the perspective that the next decade of AI adoption in construction should consider emphasizing the reduction of repetitive digital cognitive labor rather than focusing solely on the expansion of field robotics and sensor-based monitoring. To establish this position, the authors conducted a deep literature review to identify current academic research trends in the domain of AI in construction and to synthesize the dominant research narratives, which revealed how sparse the research remains in the domains of documentation and generative AI. To ground this perspective in evidence, the authors conducted a structured literature review following a systematic query–filter–synthesis process. Using Scopus as the primary database, the search combined keywords related to artificial intelligence, machine learning, deep learning, and construction, spanning the years 2004–2025. The initial dataset of 24,978 publications was refined by removing book chapters, review papers, and non-English records. Metadata-based filtering and manual inspection were then used to identify core application areas and group them into thematic domains such as safety and monitoring, sustainability and environmental performance, digital and design technologies, and administrative and management processes. Quantitative patterns from this dataset were interpreted through a conceptual lens contrasting physical automation (sensor- and robotics-driven AI) with cognitive automation (document- and language-driven AI). This framework anchors the argument developed throughout the paper shifting the discourse from sensor-based perception toward document-native cognition in construction AI.
Building upon this foundation, the authors developed and present a case study of an AI-powered submittals extraction and conversational retrieval tool. The system automates the extraction of submittal requirements from technical specifications, generates transmittal-ready Excel files and cover letter Word documents, and enables project teams to query the extracted content through a chatbot interface. Hosted on Azure and deployed via a Streamlit front end, the developed tool allows superintendents and project managers to complete substantial portions of the submittal process without manually scrolling through extensive PDF documents. By aligning with conversational AI paradigms that professionals are increasingly accustomed to through platforms such as ChatGPT, Gemini, and Claude, the developed tool demonstrates a practical, near-term pathway for lowering cognitive load and improving operational efficiency. Theoretically, this paper positions the evolution of AI in construction as a shift from perceptual automation, machines perceiving the physical world through sensors to cognitive automation, machines interpreting and reasoning through documents, contracts, and correspondence. This distinction highlights that the next frontier of innovation is not new sensors but new forms of reasoning that augment human expertise rather than replace it.
2 Artificial intelligence (AI) in construction: dominant research narratives
In this section, the authors conduct a detailed review of AI applications in construction spanning the past two decades, beginning with the core domains of application, moving to the most dominant sensor-driven approaches for jobsite data collection and analytics, and then synthesizing the widely adopted AI/ML/DL concepts. Building on this review, the authors also present a mini bibliometric study of the dominant research narratives across these two decades and conclude by examining recent advancements in generative AI along with the opportunities and challenges these emerging methods present for the industry.
2.1 Core domains of AI application in construction
To situate artificial intelligence (AI) within the construction domain, it is useful to begin with its core subfields of robotics, computer vision, and machine learning, which have driven much of the early innovation. Initial applications focused on vision-centric and automation tools, including CAD-assisted workflows for prefabrication, 2D/3D photogrammetry for progress monitoring, VR/AR for design visualization and coordination, and robotic equipment for repetitive, high-precision tasks (Xue et al., 2021; Bock, 2015; Alwisy et al., 2012; Lee et al., 2023). As machine learning (ML) advanced through the development of high-performance GPUs and scalable training frameworks (Baduge et al., 2022), its role expanded from pattern recognition toward predictive and optimization-driven solutions in scheduling, design optimization, risk monitoring, resource allocation, waste reduction, and BIM-integrated data management (Al-Sinan et al., 2024; Wu et al., 2023; Jianfeng et al., 2020; Gondia et al., 2020; Selvam et al., 2024; Lu et al., 2021; Mostafa et al., 2023). This progression reflects a shift from statistical models to machine learning–based prediction and, more recently, to deep learning pipelines embedded in BIM and IoT ecosystems (Xu et al., 2021; Shayboun et al., 2019). Academic surveys confirm the widespread use of these techniques across design optimization, safety analytics, and cost forecasting (Baduge et al., 2022; Pan and Zhang, 2021; Akinosho et al., 2020; Abioye et al., 2021; Chen and Ying, 2022; Datta et al., 2024; Regona et al., 2022).
Conceptually, this trajectory reflects a research paradigm centered on perceptual automation, where AI interprets the physical environment through sensors and imagery, rather than cognitive automation, where AI engages with textual, semantic, and procedural information. While these domains form the technical backbone of AI in construction, this paper argues that future advances will depend on extending AI’s role beyond perception and control toward reasoning and interpretation, particularly within documentation and decision support, which remain comparatively underexplored.
2.2 Sensor-driven applications for jobsite data collection and analytics
The past decade has been characterized by the integration of sensor technologies including IoT devices, RGB/thermal/3D cameras, and LiDAR with machine learning (ML) and deep learning (DL) approaches to enable real-time monitoring and partial automation of field processes (Rangasamy and Yang, 2024; Jiang et al., 2022). Vision–LiDAR pipelines have advanced as-built reconstruction and automated schedule updating (Xue et al., 2021; Abbas et al., 2020), while vision-based optimization methods are increasingly applied to track materials, equipment, and labor, allowing for the diagnosis of idle time and the identification of congestion hot spots (Xiao and Kang, 2021; Park et al., 2011). Quality control has similarly shifted toward vision-driven inspection systems capable of defect detection and specification compliance verification (Guo et al., 2020; Kalasapudi et al., 2017). Safety has emerged as a flagship domain of research and application, evolving from early pattern-recognition risk models (Gheisari and Esmaeili, 2019; Park et al., 2017; Rao et al., 2022; Zhang et al., 2017) to pose/interaction analysis of workers and equipment for near-miss prediction (Lee et al., 2022; Man et al., 2022), and more recently to DL-based detection of personal protective equipment (PPE) violations and fall hazards (Akinsemoyin et al., 2023; Liu et al., 2022). In parallel, the combination of photogrammetry and LiDAR has facilitated large-scale as-built modeling for visualization, facility management, and quality assessment (Alidoost et al., 2019; Bacharidis et al., 2020; Barranquero et al., 2023), while augmented and virtual reality (AR/VR) technologies have matured into integrated workflows for coordination, training, and defect tracking, often linked to building information modeling (BIM) platforms (Chen et al., 2021; Ratajczak et al., 2019; Salem et al., 2020; Yigitbas et al., 2023; Seyman Guray et al., 2023). This trajectory exemplifies how AI research in construction has been anchored in perceptual intelligence that is, teaching machines to see, sense, and measure the physical jobsite with precision. Such approaches have yielded measurable improvements in safety and productivity but remain fundamentally data-driven rather than knowledge-driven. As this paper argues, the next stage of progress lies in cognitive intelligence, where AI interprets meaning and intent within documents and communications rather than just detecting objects and patterns in space. Recognizing this distinction between perceptual and cognitive automation is essential to balance continued innovation in field sensing with equally impactful advances in documentation-centric reasoning.
2.3 Dominant research narratives in two decades of AI applications in construction
In this paper, the authors conducted a comprehensive Scopus sweep in early 2025 (N = 24,978), which revealed that publications were concentrated in a few large clusters (Figure 1). To simplify interpretation, some categories were consolidated at a higher level: Prefabricated and Modular Construction (468) and Construction 4.0 (Bond-Taylor et al., 2021) were combined into Digital and Design Technologies, while Building Climate Control (151) was merged into Sustainability and Environmental Performance. This regrouping reflects methodological and topical similarities, such as shared emphases on digital workflows, offsite fabrication, energy efficiency, and environmental modeling.
Figure 1. Grouped thematic distribution of AI research in building construction (2004–2025).
The authors observed that the largest cluster is Sustainability and Environmental Performance, accounting for 8,096 papers, followed by Sensing, Monitoring and Safety with 6,343 papers and Digital and Design Technologies with 6,145 papers. A fourth major area, Cost, Scheduling and Productivity, includes 4,179 papers. Strikingly, Administrative and Management Processes remains underrepresented at only 215 papers, which is less than one percent of the total, highlighting the limited attention given to AI applications for automating repetitive cognitive and documentation tasks. The analysis reflected a strong concentration in domains where data is abundant and key performance indicators are standardized, such as sensor-rich jobsite monitoring (vision, LiDAR, IoT), model-centric digitalization (BIM, digital twins, optimization), sustainability analytics (energy, carbon, life cycle assessment), and predictive decision support for cost, schedule, and productivity. In contrast, low-volume topics are often overshadowed by these dominant themes. This distribution reveals a structural bias in the research paradigm: construction AI has evolved around data-driven physical intelligence, privileging tasks where measurable input–output relationships can be optimized through abundant sensor data. Areas requiring semantic interpretation, such as contracts, submittals, and specifications, remain peripheral because they demand context-sensitive reasoning rather than numerical prediction. This imbalance underscores the need to expand AI research beyond perception and control toward cognitive automation, where algorithms augment not replace human interpretation in documentation, compliance, and communication workflows.
The developed figure illustrates how AI research in construction has evolved unevenly across domains, revealing not only topical diversity but also differences in underlying data cognition. The authors interpret this distribution as evidence of two distinct paradigms driving progress. Sensor-driven studies predominantly operate on quantitative, continuous data streams such as imagery, LiDAR scans, and IoT sensor feeds, focusing on perception, detection, and prediction within the physical environment. In contrast, document-driven AI engages with qualitative and context-rich text data, including specifications, contracts, and submittals that encode professional reasoning and intent. This contrast highlights a fundamental cognitive divide: while sensor-based AI advances automation through measurement and control, document-based AI advances interpretation, reasoning, and traceability. The authors view this divergence as a key factor explaining why AI in construction has matured rapidly in measurable, data-abundant domains, yet remains underdeveloped in documentation-centric workflows that require contextual understanding and domain judgment.
2.4 Generative AI in construction: opportunities and challenges going forward
Recent literature has expanded the discussion of AI in construction to include generative AI (gen-AI) models such as ChatGPT, Google Gemini, and other large language models (LLMs). Academic reviews highlight gen-AI’s potential to enhance design ideation, recommend eco-friendly materials, optimize building parameters, propose innovative technologies, and solve complex engineering challenges through data-driven insights (Rane, 2023; Ghimire et al., 2023; Regona et al., 2022). Beyond design, gen-AI is also envisioned as a tool for project execution, enabling predictive analytics, safety enhancement, risk management, and even urban-scale planning when linked with mapping and infrastructure datasets (Abioye et al., 2021; Liladhar Rane et al., 2024). Importantly, the natural language capabilities of these models offer a pathway to make complex contracts and specifications more accessible to non-experts, lowering barriers to understanding (Gao et al., 2025).
Several studies frame these developments within broader Industry 4.0 and post-COVID priorities, particularly sustainability and digitalization (Ghimire et al., 2023). Identified opportunity areas include digital twins for predictive maintenance, robotics and autonomous construction, cost and schedule optimization, and real-time hazard detection using deep learning (Ghimire et al., 2024; Ghimire et al., 2023; Khan et al., 2025; Informatics et al., 2024). For example, BERT-based approaches have been demonstrated for automatic detection of contractual risk clauses, and “Robo-GPT” has been proposed to leverage GPT reasoning for sequence planning in robot-based assembly systems (Baduge et al., 2022). From a technical standpoint, gen-AI in construction is underpinned by architectures such as generative adversarial networks (GANs), variational autoencoders (VAEs), autoregressive models, diffusion models, and flow-based models (Bond-Taylor et al., 2021). These methods allow outputs across text, images, video, and audio, supporting integration with heterogeneous construction datasets that combine documents, images, and sensor-derived information (Baduge et al., 2022). Early deployments are most visible in preconstruction, including project information modeling (PIM) integration and design assistance, with growing interest in automated document management, stakeholder communication, and predictive analytics (Saka et al., 2024; Khan et al., 2025; Skibniewski et al., 2023).
Yet, despite high expectations reflected in industry surveys where more than 90 percent of practitioners reported positive views on adoption (The Access Group. theaccessgroup.com, 2024), several challenges remain. These include limited domain-specific training of base models, hallucination and accuracy issues, generalizability constraints, high computational costs, and regulatory and interpretability concerns (Krouska et al., 2025). Moreover, deployment currently favors large firms with the resources to fine-tune and host such systems, leaving small and medium-sized contractors at a disadvantage (Sawhney FRICS and Pitman, 2025). Addressing these gaps will require domain-specific fine-tuning, robust evaluation frameworks, and cost-effective deployment strategies tailored to construction’s operational realities (Martin et al., 2025). The emergence of generative AI represents a paradigm shift from perceptual automation where systems detect, classify, and predict physical patterns to cognitive automation, where models interpret, reason, and generate knowledge from textual and semantic data. While most current implementations focus on design synthesis or vision-based detection, the true opportunity lies in applying generative reasoning to construction documentation, submittals, and communication flows areas where human expertise remains central. Generative AI’s value, therefore, is not in replacing human intelligence but in extending it, helping professionals interpret unstructured data, generate consistent documentation, and bridge communication gaps across teams. Building on these observations, several researchable directions emerge:
1. How can trust and reliability be quantified in domain-specific generative AI systems?
2. What multimodal architectures best fuse textual, visual, and numerical construction data for document understanding?
3. How can cost, governance, and performance trade-offs be modeled to support scalable deployment in small and medium-sized firms?
4. What evaluation frameworks are needed to benchmark hallucination control and evidence traceability in generative construction AI?
These questions reframe existing challenges into a structured research agenda, marking a pathway for advancing generative AI from experimental prototypes toward validated, domain-grounded tools for the construction industry.
2.5 The overlooked frontier: documentation automation
Construction projects generate extensive documentation across both pre-award and post-award phases, including design drawings, quantity takeoffs, technical specifications, contracts, submittals, requests for information (RFIs), change orders, daily reports, payroll, and compliance records (Project Management Institute, 2021). Most of these workflows still rely on manual copying, cross-checking, and routing among stakeholders, making them labor-intensive and error-prone (Artificio. artificio.ai, 2024; Jamieson et al., 2025). The authors highlight that this remains one of the least explored frontiers for AI, despite its centrality to project administration.
Recent advances point to a practical, document-native path forward. Computer vision models such as YOLO and Faster R-CNN have been applied to detect symbols and assemblies in two-dimensional drawings, particularly in MEP, HVAC, and electrical schematics, accelerating automatic quantity takeoff without displacing estimators. Supervised machine learning methods, ranging from regression and tree ensembles to deep neural networks, have been trained on historical bids and outcomes to improve early cost and duration forecasts while standardizing repetitive preconstruction packet assembly (Jamieson et al., 2025). Natural language processing (NLP)-based systems are being piloted to parse lengthy specifications, surface missing or inconsistent requirements, and assist with submittal compliance checks (Ding et al., 2022). In parallel, submittal and document management is shifting from manual copy-paste to automated extraction and structured routing, with features now appearing in commercial platforms such as AutoSpecs and Procore (Autodesk. construction.autodesk.com, 2025; Datagrid Team. datagrid.com, 2025; Smartbarrel Team. smart barrel.io, 2025). Complementary progress has been made in back-office automation. Smart capture and rule-based reconciliation increasingly support timekeeping, payroll, and compliance reporting, reducing error rates and cycle times (Smartbarrel Team. smart barrel.io, 2025). Yet adoption of documentation-centric automation still lags behind more mature field-facing AI such as robotics, BIM, and sensor-driven monitoring. Barriers include the messy nature of inputs (scans, mixed templates), privacy and contractual constraints on real project data, integration frictions with entrenched tools, and the cost and expertise required to fine-tune and govern AI models.
The authors argue that these challenges can be mitigated through lightweight pipelines that combine optical character recognition (OCR), regular expressions, and retrieval-augmented language models. This approach preserves familiar outputs such as Excel, Word, and PDF, while keeping humans in the loop where professional judgment is essential (Franzia. AI Meets Human Intelligence aimhi.ai, 2024; Amoah, 2025; Jamieson et al., 2025; Ren and Kim, 2025). Literature consistently supports the feasibility of such approaches, documenting the rise of learning-based methods across the project lifecycle (Baduge et al., 2022; Pan and Zhang, 2021; Akinosho et al., 2020; Abioye et al., 2021; Chen and Ying, 2022; Datta et al., 2024; Regona et al., 2022). Documentation automation marks a transition from physical to cognitive intelligence in construction where AI no longer observes or measures the jobsite but interprets and structures the text-based reasoning that drives it. Building on these insights, this paper advances a pragmatic path forward: document-native AI pipelines that produce outputs in formats professionals already use, integrate human oversight at critical junctures, and free construction teams to focus on higher-value site-level decisions. The case study that follows operationalizes this perspective through submittal automation, demonstrating how specification-driven extraction paired with conversational access can reallocate professional time toward review and decision-making.
3 Case study: submittals automation tool
3.1 Motivation and perspective
Building on the earlier discussion of generative AI opportunities and the overlooked potential of documentation automation, this paper operationalizes that perspective by addressing a persistent bottleneck in construction administration: extracting and packaging submittal requirements from specifications that often span 1,000 to 1,500 pages. In many organizations this remains a manual sequence of reading, annotating, copying into spreadsheets, and assembling transmittals. The process is slow and error prone, diverting superintendents and project engineers from higher value field work. To address this, the authors developed a submittals automation tool that reframes the workflow as a conversational, document-native pipeline. The tool preserves familiar outputs in Excel, Word, and PDF while significantly reducing keystrokes, navigation, and search time. The case serves as an applied illustration of the broader argument advanced in this paper: that the near-term value of AI in construction lies in reducing repetitive cognitive labor and improving decision efficiency.
3.2 Principal architecture
The developed system functions as an automated document intelligence pipeline designed for field adoption. It ingests technical specification PDFs and transforms them into structured, queryable data using a sequence of lightweight text processing and reasoning steps. PyMuPDF enables precise extraction, while regular expressions tuned to CSI-style numbering detect and segment specification sections without manual markup. These text segments are then embedded into vector representations using OpenAI models and stored in a FAISS index to enable rapid semantic retrieval. A GPT-4o-mini reasoning layer interprets the retrieved context and generates structured outputs such as section numbers, submittal requirements, and conditional notes. Standardized deliverables are then created in Excel and Word formats for immediate field use.
The architecture minimizes preprocessing complexity, avoids heavy OCR unless required, and balances precision with maintainability, principles that emphasize practical intelligence over algorithmic sophistication. Figure 2 summarizes the end-to-end logic from document upload and parsing to embedding, reasoning, and deliverable generation.
Figure 2. (a) Principle architecture and workflow schematic (b) pseudo-code for submittal extraction pipeline.
3.3 System components (end-to-end workflow)
The submittal automation system operates as a structured document intelligence pipeline composed of sequential yet modular components. Figure 2a illustrates the overall workflow from document upload and parsing to embedding, retrieval, reasoning, and export, while Figure 2b complements this by providing simplified pseudo-code that captures the logic of each stage in a concise, programmatic form. The process begins with document ingestion. Users upload specification PDFs through the Streamlit interface, which processes files in memory to minimize latency and avoid intermediate storage. Parsing and extraction are handled by PyMuPDF, which supports both partial and full document extraction. Regular expressions (detailed in Table 1) derived from CSI SectionFormat and PageFormat principles detect section and subsection numbering patterns (for example, “07 21 00” or “072100”), ensuring reliable segmentation across document types.
Table 1. Representative regex patterns for CSI-style section detection, derived from CSI SectionFormat and page format specification-writing principles.
After parsing, text segments are processed through a recursive character splitter that generates overlapping chunks for embedding. Each segment is converted into a numerical vector using the OpenAI text-embedding-ada-002 model. These vectors, stored in a FAISS IndexFlatL2, allow for rapid similarity searches during retrieval. The embeddings are designed to capture contextual meaning from textual data while maintaining interpretability and computational efficiency. Only textual content extracted directly from the specification is embedded scanned images and drawings are excluded to maintain accuracy and minimize preprocessing requirements.
Once embedded, the retrieval component activates when a user query (such as “List all submittals under 07 21 00”) is issued. Relevant text chunks are retrieved from the FAISS index and passed to a GPT-4o-mini reasoning layer, which structures them into rows containing section numbers, submittal requirements, and conditional notes. These outputs are then exported automatically into Excel or Word formats using openpyxl and python-docx libraries, aligning with conventional submittal workflows and preserving user familiarity. This structure balances interpretability and automation, keeping human checkpoints where necessary while allowing the AI to manage the bulk of text interpretation and formatting.
3.4 Design, deployment, usability, and roadmap
The submittals automation tool was implemented with a deliberate focus on adoption and field usability. Inputs remain in native PDF formats, outputs in Excel and Word, and interaction is conversational rather than procedural, an approach designed to reduce digital repetition while maintaining expert oversight. Regex-guided sectioning ensures structural integrity across specification numbering, while retrieval-augmented generation constrains responses to specification-grounded text for transparency and traceability.
Field testing with industry collaborators confirmed that the tool reduced manual processing time dramatically, tasks that previously took hours could now be completed in minutes. Practitioners described outputs as “complete,” “consistent,” and “a massive value add,” emphasizing improvements in speed and trust. These gains allowed project engineers to reallocate time from administrative work to higher-value coordination. User feedback also highlighted opportunities for refinement, including consolidated Excel exports, default prefilled cover letters, and faster conversational retrieval. These insights inform the current development roadmap, which includes (i) unified output options, (ii) preconfigured transmittal templates, and (iii) improved responsiveness for follow-up queries. Longer-term goals focus on agentic orchestration, automated routing of submittal packages, reviewer notifications, and tighter integration with Microsoft Teams for collaboration and tracking.
Collectively, these refinements reflect how iterative, feedback-driven development can evolve document-native AI from a functional prototype into a deployable and sustainable construction administration solution. While the tool demonstrated measurable efficiency gains, its effectiveness still depends on human oversight. Some misclassifications or redundant entries required expert review, highlighting that current AI systems augment rather than replace professional judgment. These moments of human–AI interaction where engineers verify extracted content or correct minor drift reaffirm the value of keeping humans in the loop. Rather than viewing such interventions as failures, they illustrate a balanced model of collaboration, where cognitive automation supports, but does not substitute, domain expertise.
3.5 Technical performance evaluation
The prototype was benchmarked on ten construction specification documents ranging from 800 to 1,500 pages. Average end-to-end processing time was approximately 6–8 min per document on a standard Azure B2ms instance, with total computational costs under USD 0.10 per file. Token utilization ranged between 40k and 60k per run, demonstrating that lightweight retrieval-augmented pipelines can scale efficiently even for large text-heavy specifications. Testing revealed two main error types. The first, semantic drift, occurred when similar submittal phrases across adjacent paragraphs generated duplicate entries. The second, context truncation, appeared when lengthy clauses exceeded the token window, omitting secondary conditions. Both were mitigated through smaller chunk sizes and adaptive prompt conditioning. These observations provide a practical baseline for evaluating future improvements in document-native AI, including context retention and semantic clustering for redundancy control.
Compared with existing commercial tools such as AutoSpecs and RAGflow, which offer robust, turnkey automation for document management, the proposed framework takes a distinct approach. It prioritizes transparency, customizability, and research reproducibility rather than proprietary encapsulation. Each stage of the workflow: parsing, embedding, retrieval, and reasoning is openly defined and modifiable, enabling researchers and practitioners to study interpretability, evaluate domain adaptation, and extend the pipeline for academic experimentation. Furthermore, while commercial products emphasize enterprise-scale automation and tightly integrated SaaS ecosystems, the proposed framework emphasizes accessibility and explainability, producing interpretable outputs directly in Excel and Word formats aligned with submittal workflows. To further clarify these distinctions, Table 2 summarizes the major differences between the proposed system and two widely cited commercial platforms. The comparison highlights how this framework is positioned as a research-grade, transparent, and extensible alternative, suitable for education, benchmarking, and methodological advancement within construction informatics.
Table 2. Comparison between the proposed system and commercial platforms.
4 Conclusion: adoption challenges and future directions
This perspective paper, grounded in collaboration with an industry partner during testing of the developed submittals automation tool, identifies three recurring adoption challenges even under low training requirements compared with traditional software rollouts. First, workflows and local conventions are highly heterogeneous, meaning any AI pipeline must remain configurable while preserving human sign-off to maintain trust. Second, multimodal limits persist, as current retrieval-augmented language models perform best on text, whereas drawings still demand robust vision models for reliable comprehension. Third, operational considerations such as cost, scale, and governance require efficient chunking, caching, and vector storage, with proactive planning for compute budgets, data privacy, and auditability.
Theoretical Priorities: Future research should examine how cognitive automation reshapes human–AI collaboration in documentation tasks. This includes exploring models of AI trust, cognitive load, and explainability to better understand how professionals interpret and validate AI-generated content.
Methodological Directions: From a methodological perspective, progress depends on developing retrieval-first, evaluation-driven pipelines. Priorities include establishing benchmarks for reliability, multimodal document understanding, and hallucination control, supported by shared evaluation harnesses and regression testing on held-out specifications.
Applied Pathways: At the application level, emphasis should shift toward scalable and agentic orchestration. Key steps include integrating submittal automation with collaboration environments (e.g., BIM, Teams, or CDEs), automating transmittal generation and reviewer routing with human checkpoints, and measuring ROI through time-on-task and error-reduction studies.
In summary, the authors’ real-world testing demonstrates that conversational, document-native AI can reallocate cognitive labor today by keeping inputs and outputs familiar, grounding reasoning in retrieved evidence, and preserving expert judgment. Moving forward, the convergence of theoretical insight, robust methods, and enterprise-ready applications will determine whether documentation automation evolves from a useful add-on to a baseline capability in construction administration.
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Author contributions
VK: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Writing – original draft, Writing – review and editing. AS: Data curation, Software, Visualization, Writing – review and editing. SG: Data curation, Software, Visualization, Writing – review and editing. CT: Conceptualization, Formal Analysis, Funding acquisition, Resources, Supervision, Writing – review and editing.
Funding
The authors declare that financial support was received for the research and/or publication of this article. This research was supported through a collaborative initiative between the University of North Florida (UNF) and Petticoat Schmitt Civil Contractors as part of the project titled “AI-Driven Business Intelligence and Operational Automation for Petticoat Schmitt.” The funding enabled the advancement of AI applications for construction process automation and the academic mentoring of graduate research students at UNF.
Conflict of interest
Author CT was employed by Petticoat-Schmitt Civil Contractors.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors declare that this study received funding from Petticoat Schmitt Civil Contractors. The funder was involved in identifying the research need for automating the submittal process and in facilitating access to their submittal database. The funder had no role in the analysis, interpretation of results, or the decision to submit the article for publication.
Generative AI statement
The authors declare that Generative AI was used in the creation of this manuscript. The writing in this manuscript is entirely original and authored by the listed contributors. Generative AI tools were used solely to assist with grammar refinement, consistency, and maintaining clarity of flow. Importantly, generative AI also forms part of the developed methodology described in this paper, where it is explicitly integrated into the submittals automation tool as a core component of the system architecture.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
Abbas, R., Westling, F. A., Skinner, C., Hanus-Smith, M., Harris, A., and Kirchner, N. (2020). “BuiltView: integrating LiDAR and BIM for real-time quality control of construction projects,” in 37th International Symposium on Automation and Robotics in Construction (Kitakyushu, Japan: IAARC Publications), 233–239.
Abioye, S. O., Oyedele, L. O., Akanbi, L., Ajayi, A., Davila Delgado, J. M., Bilal, M., et al. (2021). Artificial intelligence in the construction industry: a review of present status, opportunities and future challenges. J. Build. Eng. 44, 103299. doi:10.1016/j.jobe.2021.103299
Akinosho, T. D., Oyedele, L. O., Bilal, M., Ajayi, A. O., Delgado, M. D., Akinade, O. O., et al. (2020). Deep learning in the construction industry: a review of present status and future innovations. J. Build. Eng. 32, 101827. doi:10.1016/j.jobe.2020.101827
Akinsemoyin, A., Awolusi, I., Chakraborty, D., Al-Bayati, A. J., and Akanmu, A. (2023). Unmanned aerial systems and deep learning for safety and health activity monitoring on construction sites. Sensors 23 (15) 6690. doi:10.3390/s23156690
Al-Sinan, M. A., Bubshait, A. A., and Aljaroudi, Z. (2024). Generation of construction scheduling through machine learning and BIM: a blueprint. Buildings 14 (4), 934. doi:10.3390/buildings14040934
Alidoost, F., Arefi, H., and Tombari, F. (2019). 2D Image-To-3D model: knowledge-based 3D building reconstruction (3DBR) using single aerial images and convolutional neural networks (CNNs). Remote Sens. 11 (19), 2219. doi:10.3390/rs11192219
Alwisy, A., Al-Hussein, M., and Al-Jibouri, S. H. (2012). “BIM approach for automated drafting and design for modular construction manufacturing,” in Computing in civil engineering (Reston, VA: American Society of Civil Engineers), 221–228. doi:10.1061/9780784412343.0028
Amoah, K. (2025). “The artificial intelligence (AI) impact on construction project management,” in Proceedings of the 42nd International Symposium on Automation and Robotics in Construction.
Aqib, M., Hamza, M., Mei, Q., and Chui, Y. H. (2025). Fine-tuning large language models and evaluating retrieval methods for improved question answering on building codes. Smart Constr. 2 (3), 1–31. doi:10.55092/sc20250021
Artificio. artificio.ai (2024). Revolutionizing construction documentation. Available online at: https://artificio.ai/blog/revolutionizing-construction-documentation?utm_source=chatgpt.com (Accessed September 15, 2025)
Autodesk. construction.autodesk.com (2025). AutoSpecs: comprehensive construction submittal logs in minutes. Available online at: https://construction.autodesk.com/tools/autospecs-construction-submittal-log/(Accessed September 15, 2025)
Bacharidis, K., Sarri, F., and Ragia, L. (2020). 3D building façade reconstruction using deep learning. ISPRS Int. J. Geo-Information 9 (5), 322. doi:10.3390/ijgi9050322
Baduge, S. K., Thilakarathna, S., Perera, J. S., Arashpour, M., Sharafi, P., Teodosio, B., et al. (2022). Artificial intelligence and smart vision for building and construction 4.0: machine and deep learning methods and applications. Autom. Constr. 141, 104440. doi:10.1016/j.autcon.2022.104440
Barranquero, M., Olmedo, A., Gómez, J., Tayebi, A., Hellín, C. J., and Saez de Adana, F. (2023). Automatic 3D building reconstruction from OpenStreetMap and LiDAR using convolutional neural networks. Sensors 23 (5), 2444. doi:10.3390/s23052444
Bock, T. (2015). The future of construction automation: technological disruption and the upcoming ubiquity of robotics. Autom. Constr 59, 113–121. doi:10.1016/j.autcon.2015.07.022
Bond-Taylor, S., Leach, A., Long, Y., and Willcocks, C. G. (2021). Deep generative modelling: a comparative review of VAEs, GANs, normalizing flows, energy-based and autoregressive models. IEEE Trans. Pattern Anal. Mach. Intell. 44 (11), 7327–7347. doi:10.1109/tpami.2021.3116668
Buildots (2025). Reinventing construction process control: why digital innovation in process measurement and control holds the key to the future of construction.
Caldas, C. H., and Soibelman, L. (2003). Automating hierarchical document classification for construction management information systems. Autom. Constr 12 (4), 395–406. doi:10.1016/s0926-5805(03)00004-9
Chen, H. P., and Ying, K. C. (2022). Artificial intelligence in the construction industry: main development trajectories and future outlook. Appl. Sci. 12 (12), 5832. doi:10.3390/app12125832
Chen, H., Hou, L., Zhang, G., and Moon, S. (2021). Development of BIM, IoT and AR/VR technologies for fire safety and upskilling. Autom. Constr. 125, 103631. doi:10.1016/j.autcon.2021.103631
Chen, G., Alsharef, A., Ovid, A., Albert, A., and Jaselskis, E. (2025). Meet2Mitigate: an LLM-powered framework for real-time issue identification and mitigation from construction meeting discourse. Adv. Eng. Inf. 64, 103068. doi:10.1016/j.aei.2024.103068
Datagrid Team. datagrid.com (2025). AI in construction: automating Permits and Documentation | datagrid. Available online at: https://www.datagrid.com/blog/automate-permits-documentation-construction(Accessed September 15, 2025)
Datta, S. D., Islam, M., Rahman Sobuz, M. H., Ahmed, S., and Kar, M. (2024). Artificial intelligence and machine learning applications in the project lifecycle of the construction industry: a comprehensive review. Heliyon 10 (5), e26888. doi:10.1016/j.heliyon.2024.e26888
Ding, Y., Ma, J., and Luo, X. (2022). Applications of natural language processing in construction. Autom. Constr 136, 104169. doi:10.1016/j.autcon.2022.104169
Franzia. AI Meets Human Intelligence (aimhi.ai) (2024). 6 ways AI eliminates construction paperwork and prevents delays - AIMHI by eve. Available online at: https://aimhi.ai/6-ways-ai-eliminates-construction-paperwork-and-prevents-delays/ (Accessed September 15, 2025).
Gao, Y., Gan, Y., Chen, Y., and Chen, Y. (2025). Application of large language models to intelligently analyze long construction contract texts. Constr. Manag. Econ. 43 (3), 226–242. doi:10.1080/01446193.2024.2415676
Gheisari, M., and Esmaeili, B. (2019). Applications and requirements of unmanned aerial systems (UASs) for construction safety. Saf. Sci. 118, 230–240. doi:10.1016/j.ssci.2019.05.015
Ghimire, P., Kim, K., and Acharya, M. (2023). Generative ai in the construction industry: opportunities and challenges.
Ghimire, P., Kim, K., and Acharya, M. (2024). Opportunities and challenges of generative AI in construction industry: focusing on adoption of text-based models. Buildings 14 (1), 220. doi:10.3390/buildings14010220
Gondia, A., Siam, A., El-Dakhakhni, W., and Nassar, A. H. (2020). Machine learning algorithms for construction projects delay risk prediction. J. Constr. Eng. Manag 146 (1), 04019085. doi:10.1061/%28ASCE%29CO.1943-7862.0001736
Guo, B. H. W., Scheepbouwer, E., Yiu, T. W., and González, V. A. (2017). Overview and analysis of digital technologies for construction safety management.
Guo, J., Wang, Q., and Park, J. H. (2020). Geometric quality inspection of prefabricated MEP modules with 3D laser scanning. Autom. Constr. 111, 103053. doi:10.1016/j.autcon.2019.103053
Informatics, U., Xu, H., Omitaomu, F., Sabri, S., Zlatanova, S., Li, X., et al. (2024). Leveraging generative AI for urban digital twins: a scoping review on the autonomous generation of urban data, scenarios, designs, and 3D city models for smart city advancement. Urban Inf. 3 (1), 01–44. doi:10.1007/s44212-024-00060-w
Jacques de Sousa, L., Poças Martins, J., Sanhudo, L., and Santos Baptista, J. (2024). Automation of text document classification in the budgeting phase of the construction process: a systematic literature review. Constr. Innov. 24 (7), 292–318. doi:10.1108/CI-12-2022-0315
Jamieson, L., Moreno-Garcia, C. F., and Elyan, E. (2025). Towards fully automated processing and analysis of construction diagrams: AI-powered symbol detection. Int. J. Document Analysis Recognit. 28 (1), 71–84. doi:10.1007/s10032-024-00492-9
Jang, Y., Asce, A. M., Kim, K., Leite, F., Asce, M., Ayer, S., et al. (2021). Identifying the perception differences of emerging construction-related technologies between industry and academia to enable high levels of collaboration. J. Constr. Eng. Manag. 147 (10), 06021004. doi:10.1061/(asce)co.1943-7862.0002156
Jianfeng, Z., Yechao, J., and Fang, L. (2020). “Construction of intelligent building design system based on BIM and AI,” in 2020 5th International Conference on Smart Grid and Electrical Automation (ICSGEA) (IEEE), 277–280. Available online at: https://ieeexplore.ieee.org/document/9260391/.
Jiang, H., Zhang, L., and Lv, W. (2022). “The impact of STEM competitions on students’ career interest and persistence in STEM,” in Proceedings - 2022 4th International Conference on Computer Science and Technologies in Education, CSTE 2022 (Xi'an, China: Institute of Electrical and Electronics Engineers Inc.), 279–283.
Kalasapudi, V. S., Tang, P., and Turkan, Y. (2017). Computationally efficient change analysis of piece-wise cylindrical building elements for proactive project control. Autom. Constr. 81, 300–312. doi:10.1016/j.autcon.2017.04.001
Khan, A., Chang, S., and Chang, H. (2025). Generative AI approaches for architectural design automation. Autom. Constr. 180, 106506. doi:10.1016/j.autcon.2025.106506
Knyazeva, N., and Larin, V. (2021). Automation of repetitive tasks when designing an eco-friendly residential home. E3S Web Conf. 244, 05021. doi:10.1051/e3sconf/202124405021
Krouska, A., Troussas, C., Kardaras, D., Sgouropoulou, C., Adebayo, Y., Udoh, P., et al. (2025). Artificial intelligence in construction project management: a structured literature review of its evolution in application and future trends. Digital 5 (3), 26. doi:10.3390/digital5030026
Lee, S., Koo, B., Yang, S., Kim, J., Nam, Y., and Kim, Y. (2022). Fall-from-Height detection using deep learning based on IMU sensor data for accident prevention at construction sites. Sensors 22 (16), 6107. doi:10.3390/s22166107
Lee, J. K., Lee, S., Kim, Y. C., Kim, S., and Hong, S. W. (2023). Augmented virtual reality and 360 spatial visualization for supporting user-engaged design. J. Comput. Des. Eng. 10 (3), 1047–1059. doi:10.1093/jcde/qwad035
Liladhar Rane, N., Choudhary, S. P., and Rane, J. (2024). Artificial intelligence acceptance and implementation in construction industry: factors, current trends, and challenges. Available online at: https://ssrn.com/abstract=4841619 (Accessed September 15, 2025).
Liu, J., Luo, H., and Liu, H. (2022). Deep learning-based data analytics for safety in construction. Autom. Constr. 140, 104302. doi:10.1016/j.autcon.2022.104302
Lu, W., Lou, J., Webster, C., Xue, F., Bao, Z., and Chi, B. (2021). Estimating construction waste generation in the greater Bay area, China using machine learning. Waste Manag. 134, 78–88. doi:10.1016/j.wasman.2021.08.012
Man, Y., Tran, K. P., Hayat, A., and Morgado-Dias, F. (2022). Deep learning-based automatic safety helmet detection system for construction safety. Appl. Sci. 12 (16), 8268. doi:10.3390/app12168268
Martin, H., Asce, M., James, J., and Chadee, A. (2025). Exploring large language model AI tools in construction project risk assessment: chat GPT limitations in risk identification, mitigation strategies, and user experience. J. Constr. Eng. Manag. 151 (9), 04025119. doi:10.1061/JCEMD4.COENG-16658
Mostafa, A. L., Mohamed, M. A., Ahmed, S., and Youssef, WMMA (2023). Application of artificial intelligence tools with BIM technology in construction management: literature review. Int. J. BIM Eng. Sci. 6 (2), 39–54. doi:10.54216/ijbes.060203
Pan, Y., and Zhang, L. (2021). Roles of artificial intelligence in construction engineering and management: a critical review and future trends. Automation Constr. 122, 103517. doi:10.1016/j.autcon.2020.103517
Park, M. W., Makhmalbaf, A., and Brilakis, I. (2011). Comparative study of vision tracking methods for tracking of construction site resources. Autom. Constr. 20 (7), 905–915. doi:10.1016/j.autcon.2011.03.007
Park, J., Kim, K., and Cho, Y. K. (2017). Framework of automated construction-safety monitoring using cloud-enabled BIM and BLE Mobile tracking sensors. J. Constr. Eng. Manag. 143 (2), 05016019. doi:10.1061/%28ASCE%29CO.1943-7862.0001223
Project Management Institute (2021). PMBOK guide | A guide to the project management body of knowledge. Available online at: https://www.pmi.org/standards/pmbok (Accessed September 15, 2025).
Rane, N. (2023). Role of ChatGPT and similar generative artificial intelligence (AI) in construction industry. SSRN Electron. J. doi:10.2139/ssrn.4598258
Rangasamy, V., and Yang, J. B. (2024). The convergence of BIM, AI and IoT: reshaping the future of prefabricated construction. J. Build. Eng. 84, 108606. doi:10.1016/j.jobe.2024.108606
Rao, A. S., Radanovic, M., Liu, Y., Hu, S., Fang, Y., Khoshelham, K., et al. (2022). Real-time monitoring of construction sites: sensors, methods, and applications. Autom. Constr. 136, 104099. doi:10.1016/j.autcon.2021.104099
Ratajczak, J., Riedl, M., and Matt, D. T. (2019). BIM-based and AR application combined with location-based management system for the improvement of the construction performance. Buildings 9 (5), 118. doi:10.3390/buildings9050118
Regona, M., Yigitcanlar, T., Xia, B., and Li, R. Y. M. (2022). Opportunities and adoption challenges of AI in the construction industry: a PRISMA review. J. Open Innovation Technol. Mark. Complex. 8 (1), 45. doi:10.3390/joitmc8010045
Ren, Z., and Kim, J. I. (2025). The role of AI in On-Site construction robotics: a state-of-the-art review using the sense–think–act framework. Buildings 15 (13), 2374. doi:10.3390/buildings15132374
Saka, A. B., Oyedele, L. O., Akanbi, L. A., Ganiyu, S. A., Chan, D. W. M., and Bello, S. A. (2023). Conversational artificial intelligence in the AEC industry: a review of present status, challenges and opportunities. Adv. Eng. Inf., 55, 101869, doi:10.1016/j.aei.2022.101869
Saka, A., Taiwo, R., Saka, N., Salami, B. A., Ajayi, S., Akande, K., et al. (2024). GPT models in construction industry: opportunities, limitations, and a use case validation. Dev. Built Environ. 17, 100300. doi:10.1016/j.dibe.2023.100300
Salem, T., and Dragomir, M. (2022). Options for and challenges of employing digital twins in construction management. Appl. Sci. 12 (6), 2928. doi:10.3390/app12062928
Salem, O., Samuel, I. J., and He, S. (2020). BIM and VR/AR technologies: from project development to lifecycle asset management. Proc. Int. Struct. Eng. Constr. 7 (1). doi:10.14455/isec.res.2020.7(1).aae-11
Sawhney Frics, A., and Pitman, K. (2025). RICS artificial intelligence in construction report 2025. Available online at: https://www.rics.org/news-insights/artificial-intelligence-in-construction-report.(Accessed September 15, 2025)
Selvam, G., Kamalanandhini, M., Velpandian, M., and Shah, S. (2024). Duration and resource constraint prediction models for construction projects using regression machine learning method. Eng. Constr. Archit. Manag. 32, 5743–5763. doi:10.1108/ECAM-06-2023-0582
Seyman Guray, T., and Kismet, B. (2023). VR and AR in construction management research: bibliometric and descriptive analyses. Smart Sustain. Built Environ. 12 (3), 635–659. doi:10.1108/sasbe-01-2022-0015
Shayboun, M., Kifokeris, D., and Koch, C. (2019). “Construction planning with machine learning,” in Proceedings of the 35th annual ARCOM conference. Association of researchers in construction management. 699–708.
Skibniewski, M. J., Liu, H., Lei, Y., Prieto, S. A., Mengiste, E. T., and García De Soto, B. (2023). Investigating the use of ChatGPT for the scheduling of construction projects. Buildings 13 (4), 857. doi:10.3390/buildings13040857
Smartbarrel Team. smartbarrel.io (2025). Discover 7 top AI tools for construction in 2025. Available online at: https://smartbarrel.io/blog/7-top-construction-ai-solutions/ (Accessed September 15, 2025).
Taneja, S., Akinci, B., Asce, M., Garrett, J. H., Soibelman, L., Ergen, E., et al. (2010). Sensing and field data capture for construction and facility operations. J. Constr. Eng. Manag. 137 (10), 870–881. doi:10.1061/(asce)co.1943-7862.0000332
The Access Group. theaccessgroup.com (2024). Insights - AI at work | the access group ANZ. Available online at: https://www.theaccessgroup.com/en-au/evo/insights/ (Accessed September 15, 2025).
Vasenev, A., Hartmann, T., and Dorée, A. G. (2014). A distributed data collection and management framework for tracking construction operations. Adv. Eng. Inf. 28 (2), 127–137. doi:10.1016/j.aei.2014.01.003
Wu, K., Mengiste, E., and García de Soto, B. (2023). “A machine learning framework for construction planning and scheduling,” in Proceedings of the creative construction conference (Abu Dhabi, United Arab Emirates: Budapest University of Technology and Economics), 391–399. Available online at: http://hdl.handle.net/10890/51311 (Accessed September 15, 2025).
Xiao, B., and Kang, S. C. (2021). Vision-based method integrating deep learning detection for tracking multiple construction machines. J. Comput. Civ. Eng. 35 (2), 04020071. doi:10.1061/%28ASCE%29CP.1943-5487.0000957
Xiao, C., Liu, Y., and Akhnoukh, A. (2018). Bibliometric review of artificial intelligence (AI) in construction engineering and management. 32-41. doi:10.1061/9780784481721.004
Xu, Y., Zhou, Y., Sekula, P., and Ding, L. (2021). Machine learning in construction: from shallow to deep learning. Dev. Built Environ. 6, 100045. doi:10.1016/j.dibe.2021.100045
Xue, J., Hou, X., and Zeng, Y. (2021). Review of image-based 3D reconstruction of building for automated construction progress monitoring. Appl. Sci. 11 (17), 7840. doi:10.3390/app11177840
Yigitbas, E., Nowosad, A., and Engels, G. (2023). Supporting construction and architectural visualization through BIM and AR/VR: a systematic literature review. 145-166. doi:10.1007/978-3-031-42283-6_8
Keywords: construction documentation automation, submittals and specifications, construction project management, generative AI, artificial intelligence in construction, human-in-the-loop
Citation: Kalasapudi VS, Seelam A, Ganupa S and Tofferi C (2026) The overlooked frontier of AI in construction: conversational, document-native automation for administrative workflows. Front. Built Environ. 11:1713342. doi: 10.3389/fbuil.2025.1713342
Received: 25 September 2025; Accepted: 24 November 2025;
Published: 02 January 2026.
Edited by:
Ci-Jyun Liang, Stony Brook University, United StatesReviewed by:
Xiaoyu Hou, Michigan Technological University, United StatesFanru Gao, Stony Brook University, United States
Copyright © 2026 Kalasapudi, Seelam, Ganupa and Tofferi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Vamsi Sai Kalasapudi, VmFtc2kua2FsYXNhcHVkaUB1bmYuZWR1