Nadeem Iqbal
Research Subjects
Nadeem Iqbal has a total of 3 co-authors. Here are the top co-authors by number of shared publications:
- Mohamed Noureldin
Department of Civil Engineering, Aalto University, Espoo, Finland
2 shared publications Khurram Shabbir
1 shared publication- Youssef Hosni1 shared publication
Latest Publications
- Open Access
- 14 October 2026
- Published by Elsevier in SSRN Electronic Journal
Multi-agent retrieval-augmented generation with layered validation for life cycle assessment and cost estimation of reinforced concrete structures
Life Cycle Assessment and Life Cycle Cost Assessment are central to sustainable structural design, yet conventional workflows are data-intensive and fragmented, limiting early-stage use. Machine learning models predict embodied carbon, energy and cost from design parameters but need structured inputs and remain isolated from technical knowledge. Large language models offer natural language access to that knowledge yet hallucinate, and conventional retrieval-augmented generation lacks task planning, deterministic tool use and output verification.This paper presents a multi-agent retrieval-augmented framework for structural sustainability assessment. A planning layer routes each query to an expert pathway grounded in a curated document corpus, a predictive pathway invoking nine trained regression models covering carbon, energy and cost for concrete, reinforcement and earthwork, a mixed pathway, or a web pathway. The predictive pathway separates language understanding from engineering computation: the language model extracts only the physical quantities stated, while a deterministic resolver applies tabulated conversions from material grade and vehicle capacity to emission, energy and cost factors. Three validation agents check evidence consistency, query coverage and factual grounding, with one bounded revision on failure.Evaluation on 51 questions scored by three independent judge models yielded a mean score of 17.33 of 18, with faithfulness 4.86 of 5. Of 23 numerical outputs, 16 fell within a 5% tolerance at a median absolute error of 1.39%, and the validation layer revised 27.4% of first-pass responses. Prediction error scales inversely with material quantity, a consequence of target normalisation rather than any engineering mechanism, bounding where predictions are reliable.
- 14 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMulti-agent retrieval-augmented generation with layered validation for life cycle assessment and cost estimation of reinforced concrete structures
Life Cycle Assessment and Life Cycle Cost Assessment are central to sustainable structural design, yet conventional workflows are data-intensive and fragmented, limiting early-stage use. Machine learning models predict embodied carbon, energy and cost from design parameters but need structured inputs and remain isolated from technical knowledge. Large language models offer natural language access to that knowledge yet hallucinate, and conventional retrieval-augmented generation lacks task planning, deterministic tool use and output verification.This paper presents a multi-agent retrieval-augmented framework for structural sustainability assessment. A planning layer routes each query to an expert pathway grounded in a curated document corpus, a predictive pathway invoking nine trained regression models covering carbon, energy and cost for concrete, reinforcement and earthwork, a mixed pathway, or a web pathway. The predictive pathway separates language understanding from engineering computation: the language model extracts only the physical quantities stated, while a deterministic resolver applies tabulated conversions from material grade and vehicle capacity to emission, energy and cost factors. Three validation agents check evidence consistency, query coverage and factual grounding, with one bounded revision on failure.Evaluation on 51 questions scored by three independent judge models yielded a mean score of 17.33 of 18, with faithfulness 4.86 of 5. Of 23 numerical outputs, 16 fell within a 5% tolerance at a median absolute error of 1.39%, and the validation layer revised 27.4% of first-pass responses. Prediction error scales inversely with material quantity, a consequence of target normalisation rather than any engineering mechanism, bounding where predictions are reliable.
- Open Access
- 15 October 2025
- Published by Elsevier in Machine Learning with Applications
- Vol. 22
A novel explainable AI-based design optimization framework to estimate sustainability and economic impacts of reinforced concrete structures
- 15 October 2025
- Published by Elsevier in Machine Learning with Applications
- Vol. 22
Open AccessA novel explainable AI-based design optimization framework to estimate sustainability and economic impacts of reinforced concrete structures