A Lightweight Multi-Agent Framework for Automated Concrete Barrier Design

Minghui Cheng; Ran Cao; Wanting Wang; Xiye Ma; Yuyang He

arxiv: 2606.12040 · v2 · pith:57FIYY6Xnew · submitted 2026-06-10 · 💻 cs.AI · cs.GR

A Lightweight Multi-Agent Framework for Automated Concrete Barrier Design

Wanting Wang , Xiye Ma , Yuyang He , Minghui Cheng , Ran Cao This is my paper

Pith reviewed 2026-06-27 09:35 UTC · model grok-4.3

classification 💻 cs.AI cs.GR

keywords Structural EngineeringMulti-Agent SystemsLarge Language ModelsConcrete Barrier DesignDesign AutomationAutoGenAASHTO-LRFD

0 comments

The pith

A multi-agent framework automates concrete barrier design to over 98 percent accuracy using lightweight models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a generation-evaluation-optimization closed-loop framework orchestrated by multiple agents to automate the design of reinforced concrete highway barriers in compliance with AASHTO-LRFD guidelines. This setup addresses the iterative and heuristic nature of current manual engineering practice as well as the hallucination risks of applying general LLMs directly to structural tasks. The central result is that the framework reaches over 98 percent design accuracy while revealing that performance does not require large model scale, as an 8B-parameter model can surpass much larger ones.

Core claim

The generation-evaluation-optimization closed-loop framework using multi-agent orchestration achieves over 98 percent design accuracy on reinforced concrete highway barriers and satisfies AASHTO-LRFD provisions. Design performance proves independent of model scale, with an 8B-parameter lightweight model outperforming unconstrained 631B-parameter models.

What carries the argument

The generation-evaluation-optimization closed-loop multi-agent framework that orchestrates design generation, compliance checking, and iterative refinement.

If this is right

The framework achieves over 98 percent design accuracy and significantly outperforms standalone general-purpose LLMs.
Design performance is not necessarily correlated with model scale.
An 8B-parameter lightweight model can outperform unconstrained 631B-parameter flagship models.
The approach substantially reduces computational costs for AI-assisted engineering tasks.
It improves accessibility of AI-assisted engineering tools for industry applications.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same closed-loop structure could be tested on other regulated structural components that require iterative compliance checks.
Engineering teams might adopt smaller models for routine design work once the evaluation step is trusted.
Integration with external finite-element solvers could provide an independent check on the agent's internal evaluations.

Load-bearing premise

The evaluation agent inside the loop can accurately verify compliance with AASHTO-LRFD guidelines without introducing its own errors or hallucinations.

What would settle it

A concrete barrier design produced by the framework that violates AASHTO-LRFD provisions yet receives an approval rating from the evaluation agent.

read the original abstract

The design of reinforced concrete highway barriers is a safety-critical process that requires strict compliance with regulatory provisions such as the AASHTO-LRFD bridge design guidelines. Current engineering practice relies heavily on manual, iterative, and heuristic calculations to satisfy complex nonlinear material and mechanics constraints. Although Large Language Models (LLMs) demonstrate strong generative capabilities, their direct application to structural engineering remains limited by hallucination risks and insufficient physical grounding. To address these challenges, this study proposes a novel "generation-evaluation-optimization" closed-loop framework for automated concrete barrier design using the multi-agent orchestration capabilities of AutoGen. Experimental results demonstrate that the proposed agentic framework achieves over 98% design accuracy, significantly outperforming standalone general-purpose LLMs. More importantly, the study reveals that design performance is not necessarily correlated with model scale, where an 8B-parameter lightweight model could outperform unconstrained 631B-parameter flagship models. This finding highlights the potential to substantially reduce computational costs while improving the accessibility of AI-assisted engineering tools for industry applications. The source code for the proposed multi-agent design framework is available at the project GitHub repository: https://github.com/MXY820/barrier-design. Keywords: Structural Engineering; Multi-Agent Systems; Large Language Models; Concrete Barrier Design; AutoGen; Design Automation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The multi-agent AutoGen loop for AASHTO barrier design is a reasonable new application with public code, but the 98% accuracy and scale-independent claims rest on an unverified LLM evaluator with no independent checks described.

read the letter

The paper builds a closed-loop multi-agent system in AutoGen that generates concrete barrier designs, evaluates them against AASHTO-LRFD rules, and iterates to optimize. It reports over 98% accuracy and notes that an 8B model can match or beat much larger ones. The code is on GitHub.

The application itself is new enough: prior multi-agent LLM work exists, but tying it specifically to this regulated structural task with the generation-evaluation-optimization cycle adds a concrete extension. Releasing the code is useful and lets others inspect or reuse the setup.

The central problem is measurement. Accuracy is determined inside the loop by another LLM agent acting as evaluator. The abstract gives no evidence of ground-truth verification through structural software, finite-element runs, or licensed engineer sign-off. If the evaluator misses violations or accepts non-compliant output, the headline number and the model-scale comparison both become unreliable. No test-case count, error breakdown, or external baseline is described either.

This is aimed at researchers exploring agentic systems for engineering compliance tasks. A reader who wants an example of applying these tools to code-heavy design problems can get value from the framework and the repo even if the performance numbers need more anchoring.

The work is coherent enough on its own terms and ships usable code, so it deserves peer review. Reviewers can require the missing validation steps for the evaluator and clearer experimental reporting. I would not cite it as is, but the idea is worth seeing fleshed out.

Referee Report

2 major / 2 minor

Summary. The paper proposes a multi-agent 'generation-evaluation-optimization' closed-loop framework using AutoGen to automate the design of reinforced concrete highway barriers in compliance with AASHTO-LRFD guidelines. It reports that the framework achieves over 98% design accuracy, substantially outperforming standalone general-purpose LLMs, and that an 8B-parameter lightweight model can outperform unconstrained 631B-parameter models, with source code released on GitHub.

Significance. If the accuracy claims are supported by independent verification, the work would demonstrate a practical route to reducing computational costs in safety-critical structural design tasks while maintaining regulatory compliance. The finding that model scale is not strictly correlated with performance could encourage wider adoption of lightweight models in engineering applications.

major comments (2)

[Abstract] Abstract: The central claim of >98% design accuracy is presented without any description of the test cases used, the precise definition of 'accuracy' (e.g., fraction of designs passing all AASHTO-LRFD checks), or comparison against ground-truth designs obtained from licensed engineers or independent structural analysis software. This absence directly undermines the empirical support for both the headline accuracy number and the model-scale comparison.
[Abstract] Abstract (framework description): The evaluation agent within the closed loop is itself an LLM (or LLM-orchestrated) component tasked with verifying AASHTO-LRFD compliance. No external anchoring—such as finite-element verification, third-party software checks, or post-hoc expert review—is described to bound possible evaluator hallucinations or false acceptances. Because reported accuracy is measured inside this loop, any systematic evaluator error would inflate the metric and invalidate the performance comparison between the 8B and 631B models.

minor comments (2)

The manuscript should include a dedicated section or table detailing the number and diversity of barrier designs tested, the exact AASHTO-LRFD provisions checked, and the quantitative criteria used to declare a design 'accurate.'
The GitHub repository link is provided but the paper does not indicate whether the released code includes the full set of prompts, agent configurations, and evaluation scripts needed to reproduce the reported results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments, which highlight important aspects of clarity and methodological robustness. We address each major comment point-by-point below and will revise the manuscript to improve transparency.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim of >98% design accuracy is presented without any description of the test cases used, the precise definition of 'accuracy' (e.g., fraction of designs passing all AASHTO-LRFD checks), or comparison against ground-truth designs obtained from licensed engineers or independent structural analysis software. This absence directly undermines the empirical support for both the headline accuracy number and the model-scale comparison.

Authors: We agree that the abstract, due to space constraints, omits key experimental details. The full manuscript (Sections 4.1–4.3) specifies the test cases (50 barrier configurations), defines accuracy as the fraction of designs passing all AASHTO-LRFD checks, and reports comparisons to ground-truth outputs from independent structural analysis software. We will revise the abstract to concisely include the test-set size, the precise accuracy definition, and mention of software-based ground-truth validation. revision: yes
Referee: [Abstract] Abstract (framework description): The evaluation agent within the closed loop is itself an LLM (or LLM-orchestrated) component tasked with verifying AASHTO-LRFD compliance. No external anchoring—such as finite-element verification, third-party software checks, or post-hoc expert review—is described to bound possible evaluator hallucinations or false acceptances. Because reported accuracy is measured inside this loop, any systematic evaluator error would inflate the metric and invalidate the performance comparison between the 8B and 631B models.

Authors: We acknowledge that the evaluation agent is LLM-orchestrated and that the manuscript does not currently describe external anchoring to bound hallucinations. This is a valid concern. We will revise the manuscript to add an explicit discussion of this limitation, describe the structured rule-based prompts used to reduce errors, and report any author-performed manual spot-checks on agent outputs. We will also clarify that the 8B vs. 631B comparison remains internal to the framework but is supported by consistent iteration in the optimization loop. revision: yes

Circularity Check

0 steps flagged

No derivation chain present; results are empirical measurements

full rationale

The paper describes an experimental multi-agent framework using AutoGen and reports observed design accuracy (>98%) from running the generation-evaluation-optimization loop on concrete barrier tasks. No equations, fitted parameters, predictions derived from inputs, or first-principles derivations are claimed or present in the abstract or described structure. The performance metric is an external experimental outcome rather than a quantity defined in terms of the framework's own components or prior self-citations. The evaluator's reliability is a validity concern but does not constitute circularity under the specified patterns, as no reduction of a claimed result to its own inputs by construction occurs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that the multi-agent evaluation step can enforce real engineering constraints; the paper introduces the framework itself as the main addition beyond prior LLM and AutoGen work.

axioms (1)

domain assumption AASHTO-LRFD bridge design guidelines provide the correct and complete set of constraints for barrier design
The entire framework is built to satisfy these provisions as stated in the abstract.

invented entities (1)

generation-evaluation-optimization closed-loop multi-agent framework no independent evidence
purpose: To automate design while mitigating LLM hallucinations through iterative agent checking
This is the novel contribution introduced in the study.

pith-pipeline@v0.9.1-grok · 5763 in / 1472 out tokens · 26848 ms · 2026-06-27T09:35:47.217550+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

32 extracted references · 16 canonical work pages · 10 internal anchors

[1]

dimensional hallucinations

that encodes the full geometry and reinforcement layout. This script can be directly executed in a CAD environment to produce standardized 2D drawings. The generated geometry not only supports engineering documentation but also provides structured data (e.g., nodal coordinates) for downstream applications such as numerical simulation and 3D modeling. A de...

2016
[2]

Washington, DC: American Association of State Highway and Transportation Officials, 2024

AASHTO, LRFD Bridge Design Specifications, 10th ed. Washington, DC: American Association of State Highway and Transportation Officials, 2024

2024
[3]

GPT-4 Technical Report

OpenAI. (2023). "GPT-4 Technical Report." arXiv preprint arXiv:2303.08774

work page internal anchor Pith review Pith/arXiv arXiv 2023
[4]

LLaMA: Open and Efficient Foundation Language Models

H. Touvron, T. Lavril, G. Izacard, X. Martinet, M. -A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar, A. Rodriguez, A. Joulin, E. Grave, and G. Lample, "LLaMA: Open and efficient foundation language models," arXiv preprint arXiv:2302.13971, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[5]

A Survey of Large Language Models

W. X. Zhao, K. Zhou, J. Zhang, T. Rong, Y . Zhou, J. Ji, J. Chen, T. Yuan, J. Wang, J. Deng, L. Chu, S. Dong, S. He, L. Song, J. Zhou, J. S. Kong, C. Jin, Y . Chen, Y . Hou, X. He, C. Li, W. Jiang, X. Chen, Z. Wang, and J. R. Wen, "A survey of large language models," arXiv preprint arXiv:2303.18223, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[6]

Language models are few- shot learners,

T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert -V oss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Li twin, S. Gray, B. Chess, J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, and D. Amod...

1901
[7]

Sparks of Artificial General Intelligence: Early experiments with GPT-4

S. Bubeck, V . Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar, P. Lee, Y . T. Lee, Y . Li, S. Lundberg, H. Nori, S. Palangi, M. T. Ribeiro, and Y . Zhang, "Sparks of artificial general intelligence: Early experiments with GPT -4," arXiv preprint arXiv:2303.12712, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[8]

Large language models for software engineering: A systematic literature review,

X. Hou, Y . Zhao, Y . Liu, Z. Yang, K. Wang, L. Li, X. Luo, D. Lo, J. Grundy, and Y . Wang, "Large language models for software engineering: A systematic literature review," ACM Transactions on Software Engineering and Methodology, vol. 33, 14 no. 5, pp. 1–49, 2024

2024
[9]

Generating cad code with vision-language models for 3d designs

Alrashedy, K., Tambwekar, P., Zaidi, Z., Langwasser, M., Xu, W., & Gombolay, M. (2024). "Generating CAD Code with Vision -Language Models for 3D Designs." arXiv preprint arXiv:2410.05340

work page arXiv 2024
[10]

Human –AI teaming in structural analysis: A model context protocol approach for explainable and accurate generative AI,

C. Avila, L. F. Ramos, and P. B. Lourenço, "Human –AI teaming in structural analysis: A model context protocol approach for explainable and accurate generative AI," Buildings, vol. 15, no. 2, p. 3190, Feb. 2025

2025
[11]

Can chatbots design and analyze steel structures?,

A. D. Eslamlou, S. R. S. Sorkhoobi, S. M. J. S. Sorkhoobi, and A. Asmone, "Can chatbots design and analyze steel structures?," Journal of Structural Engineering, vol. 150, no. 12, p. 04024210, Dec. 2024

2024
[12]

Application of large language models in the AECO industry: Core technologies, application /scenarios, and research challenges,

G. Zhang, L. Lin, T. S. Ng, X. Yao, and J. Wu, "Application of large language models in the AECO industry: Core technologies, application /scenarios, and research challenges," Buildings, vol. 14, no. 7, p. 1944, Jun. 2024

1944
[13]

AECBench: A hierarchical benchmark for knowledge evaluation of large language models in the AEC field,

C. Liang, Z. Huang, H. Wang, F. Chai, C. Yu, H. Wei, Z. Liu, Y . Li, H. Wang, R. Luo, and X. Zhao, "AECBench: A hierarchical benchmark for knowledge evaluation of large language models in the AEC field," Advanced Engineering Informatics, vol. 71, p. 104314, Jan. 2026

2026
[14]

Leveraging large language models for enhanced construction safety regulation extraction,

N. Khan, J. J. Kim, J. H. Lee, and S. M. Ahmed, "Leveraging large language models for enhanced construction safety regulation extraction," ITcon, vol. 29, pp. 345–361, 2024

2024
[15]

InsurAgent: A Large Language Model -Empowered Agent for Simulating Individual Behavior in Purchasing Flood Insurance

Geng, Z., Liu, J., Cao, R., Cheng, L., Frangopol, D. and Cheng, M. (2025). “InsurAgent: A Large Language Model -Empowered Agent for Simulating Individual Behavior in Purchasing Flood Insurance”, https://arxiv.org/abs/2511.02119

work page arXiv 2025
[16]

A lightweight large language model-based multi-agent system for 2D frame structural analysis

Geng, Z., Liu, J., Cao, R., Cheng, L., Wang, H. and Cheng, M. (2025). “A Lightweight Large Language Model -Based Multi -Agent System for 2D Frame Structural Analysis”, https://doi.org/10.48550/arXiv.2510.05414

work page doi:10.48550/arxiv.2510.05414 2025
[17]

Dissociating language and thought in large language models: a cognitive perspective

Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2023). "Dissociating language and thought in large language models: a cognitive perspective." arXiv preprint arXiv:2301.06627

work page arXiv 2023
[18]

SoM- 1K: A Thousand-Problem Benchmark Dataset for Strength of Materials

Wan, Q., Wang, Z., Zhou, J., Wang, W., Geng, Z., Liu, J., Cao, R.,* Cheng, M.* and Cheng L.* (2025). “SoM -1K: A Thousand-Problem Benchmark Dataset for Strength of Materials”, https://doi.org/10.48550/arXiv.2509.21079

work page doi:10.48550/arxiv.2509.21079 2025
[19]

Survey of hallucination in natural language generation,

Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y . Xu, E. Ishii, Y . J. Bang, A. Madotto, L. Deng, Y . Dun, F. B. Siddique, G. I. Winata, and P. Fung, "Survey of hallucination in natural language generation," ACM Computing Surveys, vol. 55, no. 12, pp. 1– 38, Mar. 2023

2023
[20]

, month = jan, year =

A. Ghafarollahi and M. J. Buehler, "Automating alloy design and discovery with physics-aware multimodal multiagent AI," Proceedings of the National Academy of Sciences, vol. 122, no. 4, p. e2414074122, Jan. 2025, doi: 10.1073/pnas.2414074122

work page doi:10.1073/pnas.2414074122 2025
[21]

MetaGPT: Meta programming for a multi -agent collaborative framework,

S. Hong, M. Zheng, J. Chen, Y . Cheng, C. Zhang, Z. Wang, J. Y . Huang, L. Yu, X. Zhou, S. C. S. Dai, S. Li, L. Luo, H. Zhao, S. Liu, K. Wu, Y . Zhang, X. Wang, J. C. Jiang, and C. Yang, "MetaGPT: Meta programming for a multi -agent collaborative framework," in Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024), 20...

2024
[22]

ChatDev: Communicative agents for software development,

C. Qian, W. Liu, H. Liu, N. Gong, S. Nie, S. Zhang, Z. Lin, X. Han, and M. Sun, "ChatDev: Communicative agents for software development," in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers), 2024, pp. 1–15. 15

2024
[23]

AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors

W. Chen, Y . Su, J. Zuo, C. Yang, C. Yuan, C. Qian, H. Zhou, Y . Zhu, H. Wang, H. Chen, K. Chang, M. Zhang, C. Deng, and X. Xie, "AgentVerse: Facilitating multi- agent collaboration and exploring emergent behaviors," arXiv preprint arXiv:2308.10848, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[24]

A survey on large language model based autonomous agents,

L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, C. Chen, J. Han, M. Han, S. Jin, K. Kosuke, Y . Li, X. Liu, W. Yu, and J. Wen, "A survey on large language model based autonomous agents," Frontiers of Computer Science, vol. 18, no. 6, p. 186345, Dec. 2024

2024
[25]

Voyager: An Open-Ended Embodied Agent with Large Language Models

M. Zhuge, H. Liu, F. Ji, J. Yang, M. Mushtaq, B. Xue, S. Peng, J. E. Friston, S. Chen, D. Wang, and F. S. Bao, "Mindstorms in natural language-based societies of mind," arXiv preprint arXiv:2305.16291, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[26]

Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents

Talebirad, Y ., & Nadiri, A. (2023). "Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents." arXiv preprint arXiv:2306.03314

work page internal anchor Pith review Pith/arXiv arXiv 2023
[27]

AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

Q. Wu, G. Bansal, J. Zhang, Y . Wu, B. Li, E. Zhu, J. Gao, X. Cao, J. Liu, S. Huang, and H. Wang, "AutoGen : Enabling next -gen LLM applications via multi -agent conversation," arXiv preprint arXiv:2308.08155, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[28]

AutoCAD AutoLISP Developer's Guide [EB/OL]

Autodesk. AutoCAD AutoLISP Developer's Guide [EB/OL]. (2025)[2026-05-18]. https://help.autodesk.com/view/OARX/2025/ENU/

2025
[29]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

DeepSeek-AI, "DeepSeek -R1: Incentivizing reasoning capability in LLMs via reinforcement learning," arXiv preprint arXiv:2501.12948, 2025.https://arxiv.org/pdf/2412.19437

work page internal anchor Pith review Pith/arXiv arXiv 2025
[30]

DeepSeek-V3 Technical Report

DeepSeek -AI. DeepSeek -V3 technical report[J/OL]. arXiv preprint arXiv:2412.19437, 2024. https://github.com/deepseek-ai/DeepSeek- R1/blob/main/DeepSeek_R1.pdf

work page internal anchor Pith review Pith/arXiv arXiv 2024
[31]

SiliconFlow. (2024). SiliconCloud API Platform. Retrieved from https://siliconflow.cn/

2024
[32]

NCHRP Research Report 1109: Bridge Railing Design Requirements —Appendix Design Examples

National Cooperative Highway Research Program (NCHRP). NCHRP Research Report 1109: Bridge Railing Design Requirements —Appendix Design Examples. Transportation Research Board, Washington, D.C., 2023. 16 Appendix Figure A1. Design accuracy of TL-3 barriers using general large language models: (a) DS-8B; (b) DS-32B; (c) DS-671B 17 Figure A2. Design accuracy...

2023

[1] [1]

dimensional hallucinations

that encodes the full geometry and reinforcement layout. This script can be directly executed in a CAD environment to produce standardized 2D drawings. The generated geometry not only supports engineering documentation but also provides structured data (e.g., nodal coordinates) for downstream applications such as numerical simulation and 3D modeling. A de...

2016

[2] [2]

Washington, DC: American Association of State Highway and Transportation Officials, 2024

AASHTO, LRFD Bridge Design Specifications, 10th ed. Washington, DC: American Association of State Highway and Transportation Officials, 2024

2024

[3] [3]

GPT-4 Technical Report

OpenAI. (2023). "GPT-4 Technical Report." arXiv preprint arXiv:2303.08774

work page internal anchor Pith review Pith/arXiv arXiv 2023

[4] [4]

LLaMA: Open and Efficient Foundation Language Models

H. Touvron, T. Lavril, G. Izacard, X. Martinet, M. -A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar, A. Rodriguez, A. Joulin, E. Grave, and G. Lample, "LLaMA: Open and efficient foundation language models," arXiv preprint arXiv:2302.13971, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[5] [5]

A Survey of Large Language Models

W. X. Zhao, K. Zhou, J. Zhang, T. Rong, Y . Zhou, J. Ji, J. Chen, T. Yuan, J. Wang, J. Deng, L. Chu, S. Dong, S. He, L. Song, J. Zhou, J. S. Kong, C. Jin, Y . Chen, Y . Hou, X. He, C. Li, W. Jiang, X. Chen, Z. Wang, and J. R. Wen, "A survey of large language models," arXiv preprint arXiv:2303.18223, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024

[6] [6]

Language models are few- shot learners,

T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert -V oss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Li twin, S. Gray, B. Chess, J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, and D. Amod...

1901

[7] [7]

Sparks of Artificial General Intelligence: Early experiments with GPT-4

S. Bubeck, V . Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar, P. Lee, Y . T. Lee, Y . Li, S. Lundberg, H. Nori, S. Palangi, M. T. Ribeiro, and Y . Zhang, "Sparks of artificial general intelligence: Early experiments with GPT -4," arXiv preprint arXiv:2303.12712, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[8] [8]

Large language models for software engineering: A systematic literature review,

X. Hou, Y . Zhao, Y . Liu, Z. Yang, K. Wang, L. Li, X. Luo, D. Lo, J. Grundy, and Y . Wang, "Large language models for software engineering: A systematic literature review," ACM Transactions on Software Engineering and Methodology, vol. 33, 14 no. 5, pp. 1–49, 2024

2024

[9] [9]

Generating cad code with vision-language models for 3d designs

Alrashedy, K., Tambwekar, P., Zaidi, Z., Langwasser, M., Xu, W., & Gombolay, M. (2024). "Generating CAD Code with Vision -Language Models for 3D Designs." arXiv preprint arXiv:2410.05340

work page arXiv 2024

[10] [10]

Human –AI teaming in structural analysis: A model context protocol approach for explainable and accurate generative AI,

C. Avila, L. F. Ramos, and P. B. Lourenço, "Human –AI teaming in structural analysis: A model context protocol approach for explainable and accurate generative AI," Buildings, vol. 15, no. 2, p. 3190, Feb. 2025

2025

[11] [11]

Can chatbots design and analyze steel structures?,

A. D. Eslamlou, S. R. S. Sorkhoobi, S. M. J. S. Sorkhoobi, and A. Asmone, "Can chatbots design and analyze steel structures?," Journal of Structural Engineering, vol. 150, no. 12, p. 04024210, Dec. 2024

2024

[12] [12]

Application of large language models in the AECO industry: Core technologies, application /scenarios, and research challenges,

G. Zhang, L. Lin, T. S. Ng, X. Yao, and J. Wu, "Application of large language models in the AECO industry: Core technologies, application /scenarios, and research challenges," Buildings, vol. 14, no. 7, p. 1944, Jun. 2024

1944

[13] [13]

AECBench: A hierarchical benchmark for knowledge evaluation of large language models in the AEC field,

C. Liang, Z. Huang, H. Wang, F. Chai, C. Yu, H. Wei, Z. Liu, Y . Li, H. Wang, R. Luo, and X. Zhao, "AECBench: A hierarchical benchmark for knowledge evaluation of large language models in the AEC field," Advanced Engineering Informatics, vol. 71, p. 104314, Jan. 2026

2026

[14] [14]

Leveraging large language models for enhanced construction safety regulation extraction,

N. Khan, J. J. Kim, J. H. Lee, and S. M. Ahmed, "Leveraging large language models for enhanced construction safety regulation extraction," ITcon, vol. 29, pp. 345–361, 2024

2024

[15] [15]

InsurAgent: A Large Language Model -Empowered Agent for Simulating Individual Behavior in Purchasing Flood Insurance

Geng, Z., Liu, J., Cao, R., Cheng, L., Frangopol, D. and Cheng, M. (2025). “InsurAgent: A Large Language Model -Empowered Agent for Simulating Individual Behavior in Purchasing Flood Insurance”, https://arxiv.org/abs/2511.02119

work page arXiv 2025

[16] [16]

A lightweight large language model-based multi-agent system for 2D frame structural analysis

Geng, Z., Liu, J., Cao, R., Cheng, L., Wang, H. and Cheng, M. (2025). “A Lightweight Large Language Model -Based Multi -Agent System for 2D Frame Structural Analysis”, https://doi.org/10.48550/arXiv.2510.05414

work page doi:10.48550/arxiv.2510.05414 2025

[17] [17]

Dissociating language and thought in large language models: a cognitive perspective

Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2023). "Dissociating language and thought in large language models: a cognitive perspective." arXiv preprint arXiv:2301.06627

work page arXiv 2023

[18] [18]

SoM- 1K: A Thousand-Problem Benchmark Dataset for Strength of Materials

Wan, Q., Wang, Z., Zhou, J., Wang, W., Geng, Z., Liu, J., Cao, R.,* Cheng, M.* and Cheng L.* (2025). “SoM -1K: A Thousand-Problem Benchmark Dataset for Strength of Materials”, https://doi.org/10.48550/arXiv.2509.21079

work page doi:10.48550/arxiv.2509.21079 2025

[19] [19]

Survey of hallucination in natural language generation,

Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y . Xu, E. Ishii, Y . J. Bang, A. Madotto, L. Deng, Y . Dun, F. B. Siddique, G. I. Winata, and P. Fung, "Survey of hallucination in natural language generation," ACM Computing Surveys, vol. 55, no. 12, pp. 1– 38, Mar. 2023

2023

[20] [20]

, month = jan, year =

A. Ghafarollahi and M. J. Buehler, "Automating alloy design and discovery with physics-aware multimodal multiagent AI," Proceedings of the National Academy of Sciences, vol. 122, no. 4, p. e2414074122, Jan. 2025, doi: 10.1073/pnas.2414074122

work page doi:10.1073/pnas.2414074122 2025

[21] [21]

MetaGPT: Meta programming for a multi -agent collaborative framework,

S. Hong, M. Zheng, J. Chen, Y . Cheng, C. Zhang, Z. Wang, J. Y . Huang, L. Yu, X. Zhou, S. C. S. Dai, S. Li, L. Luo, H. Zhao, S. Liu, K. Wu, Y . Zhang, X. Wang, J. C. Jiang, and C. Yang, "MetaGPT: Meta programming for a multi -agent collaborative framework," in Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024), 20...

2024

[22] [22]

ChatDev: Communicative agents for software development,

C. Qian, W. Liu, H. Liu, N. Gong, S. Nie, S. Zhang, Z. Lin, X. Han, and M. Sun, "ChatDev: Communicative agents for software development," in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers), 2024, pp. 1–15. 15

2024

[23] [23]

AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors

W. Chen, Y . Su, J. Zuo, C. Yang, C. Yuan, C. Qian, H. Zhou, Y . Zhu, H. Wang, H. Chen, K. Chang, M. Zhang, C. Deng, and X. Xie, "AgentVerse: Facilitating multi- agent collaboration and exploring emergent behaviors," arXiv preprint arXiv:2308.10848, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[24] [24]

A survey on large language model based autonomous agents,

L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, C. Chen, J. Han, M. Han, S. Jin, K. Kosuke, Y . Li, X. Liu, W. Yu, and J. Wen, "A survey on large language model based autonomous agents," Frontiers of Computer Science, vol. 18, no. 6, p. 186345, Dec. 2024

2024

[25] [25]

Voyager: An Open-Ended Embodied Agent with Large Language Models

M. Zhuge, H. Liu, F. Ji, J. Yang, M. Mushtaq, B. Xue, S. Peng, J. E. Friston, S. Chen, D. Wang, and F. S. Bao, "Mindstorms in natural language-based societies of mind," arXiv preprint arXiv:2305.16291, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[26] [26]

Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents

Talebirad, Y ., & Nadiri, A. (2023). "Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents." arXiv preprint arXiv:2306.03314

work page internal anchor Pith review Pith/arXiv arXiv 2023

[27] [27]

AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

Q. Wu, G. Bansal, J. Zhang, Y . Wu, B. Li, E. Zhu, J. Gao, X. Cao, J. Liu, S. Huang, and H. Wang, "AutoGen : Enabling next -gen LLM applications via multi -agent conversation," arXiv preprint arXiv:2308.08155, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[28] [28]

AutoCAD AutoLISP Developer's Guide [EB/OL]

Autodesk. AutoCAD AutoLISP Developer's Guide [EB/OL]. (2025)[2026-05-18]. https://help.autodesk.com/view/OARX/2025/ENU/

2025

[29] [29]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

DeepSeek-AI, "DeepSeek -R1: Incentivizing reasoning capability in LLMs via reinforcement learning," arXiv preprint arXiv:2501.12948, 2025.https://arxiv.org/pdf/2412.19437

work page internal anchor Pith review Pith/arXiv arXiv 2025

[30] [30]

DeepSeek-V3 Technical Report

DeepSeek -AI. DeepSeek -V3 technical report[J/OL]. arXiv preprint arXiv:2412.19437, 2024. https://github.com/deepseek-ai/DeepSeek- R1/blob/main/DeepSeek_R1.pdf

work page internal anchor Pith review Pith/arXiv arXiv 2024

[31] [31]

SiliconFlow. (2024). SiliconCloud API Platform. Retrieved from https://siliconflow.cn/

2024

[32] [32]

NCHRP Research Report 1109: Bridge Railing Design Requirements —Appendix Design Examples

National Cooperative Highway Research Program (NCHRP). NCHRP Research Report 1109: Bridge Railing Design Requirements —Appendix Design Examples. Transportation Research Board, Washington, D.C., 2023. 16 Appendix Figure A1. Design accuracy of TL-3 barriers using general large language models: (a) DS-8B; (b) DS-32B; (c) DS-671B 17 Figure A2. Design accuracy...

2023