pith. sign in

arxiv: 2606.26857 · v1 · pith:AVSIVL3Unew · submitted 2026-06-25 · 💻 cs.AI

LCAi: Life Cycle Assessment with big data fusion and retrieval-augmented generation-assisted interpretation

Georgios Tsironis , Juan D. Medrano-Garcia , Gonzalo Guillen-Gosalbez This is my paper

Pith reviewed 2026-06-26 05:10 UTC · model grok-4.3

classification 💻 cs.AI
keywords life cycle assessmentretrieval-augmented generationartificial intelligenceenvironmental hotspotsstrategic pathwaysbig data fusionsustainabilityhydrogen energy
0
0 comments X

The pith

A perspective-conditioned retrieval-augmented generation framework translates life cycle assessment results into actionable strategic pathways under uncertainty.

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

The paper addresses the gap in life cycle assessment where quantified environmental improvements are hard to turn into practical strategies amid technological, social, and policy uncertainty. It proposes a retrieval-augmented generation system that conditions large language models on multiple perspectives by retrieving from academic, industry, public discourse, and EU funding datasets. The method follows three steps: setting a scenario anchor for boundaries and targets, running perspective-specific micro-queries with constrained retrieval, and performing neutral synthesis of only stored outputs. This is illustrated in a case study of hydrogen replacing diesel in an Italian apple production facility. The design aims to limit hallucinations while keeping outputs diverse across domains.

Core claim

The paper claims that a perspective fusion RAG architecture operationalizes large language models for LCA interpretation through multi-perspective retrieval and controlled synthesis, enabling the translation of impact results into strategic pathways while mitigating hallucination risks and preserving cross-domain diversity, as shown in the hydrogen-enabled diesel reduction use case.

What carries the argument

The perspective fusion RAG architecture, which combines a scenario anchor, perspective-specific micro-queries with constrained retrieval from academic, industry, public discourse, and EU funding datasets, and a neutral synthesis step that integrates only ledger-stored outputs.

Load-bearing premise

That the chosen academic, industry, public discourse, and EU funding datasets, combined with perspective-specific micro-queries and a neutral synthesis step, will produce unbiased, actionable strategic pathways without missing critical uncertainties or introducing new errors.

What would settle it

Expert review of the generated pathways reveals either factual hallucinations traceable to the retrieval sources or major uncertainties and biases omitted from the neutral synthesis step.

Figures

Figures reproduced from arXiv: 2606.26857 by Georgios Tsironis, Gonzalo Guillen-Gosalbez, Juan D. Medrano-Garcia.

Figure 1
Figure 1. Figure 1: (a) Normalised annual mentions of “green hydrogen,” “renewable hydrogen,” and “renewable feedstock(s)” in Scopus-indexed literature (2000–2026). (b) Relative share of literature index for “green hydrogen,” “renewable hydrogen,” and “renewable feedstock(s)” (2000–2026). Values are indexed to the historical maximum share (max(S) = 100) for visual clarity across datasets; consequently, annual totals exceed 10… view at source ↗
Figure 2
Figure 2. Figure 2: Global distribution and digital reach of hydrogen-related companies on LinkedIn. [PITH_FULL_IMAGE:figures/full_fig_p025_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: EU-funded hydrogen research landscape via CORDIS data. [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Public discourse trends for hydrogen-related topics across social media platforms [PITH_FULL_IMAGE:figures/full_fig_p029_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Supply vector analysis of the top carbon footprint contributions for the production [PITH_FULL_IMAGE:figures/full_fig_p031_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Climate change impact breakdown for the transport activity. [PITH_FULL_IMAGE:figures/full_fig_p031_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Climate change impact breakdown for the production of apples in Italy. [PITH_FULL_IMAGE:figures/full_fig_p032_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Supply vector analysis of the top carbon footprint contributions for the production [PITH_FULL_IMAGE:figures/full_fig_p032_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of the base case production of 1 kg of apples in Italy and transport to the [PITH_FULL_IMAGE:figures/full_fig_p033_9.png] view at source ↗
read the original abstract

The interpretation phase of life cycle assessment often lacks structured mechanisms for translating quantified improvement opportunities addressing environmental hotspots into actionable strategic pathways under technological, social, and policy uncertainty. To overcome this limitation, this study introduces a perspective-conditioned retrieval-augmented generation framework for LCA interpretation, where a multi-perspective retrieval and controlled synthesis is incorporated in the artificial intelligence (AI)-assisted LCA. To operationalise large language models in LCA interpretation, a perspective fusion RAG architecture was developed, covering academic, industry, public discourse, and European union (EU) funding datasets. Our approach comprises three steps: (1) a scenario anchor defining system boundaries and decarbonization targets, (2) a set of perspective-specific micro-queries with constrained retrieval, and (3) a neutral synthesis step integrating only ledger-stored outputs without further retrieval. The framework is demonstrated through a hydrogen-enabled diesel reduction use case in an Italian apple production facility using GPT-5 nano as the reasoning model. Overall, the structured retrieval and constrained synthesis are designed to mitigate the risk of hallucination while preserving cross-domain diversity. The approach presented can support more disciplined translation of impact results into strategic pathways and opens up new avenues for the use of advanced AI tools in LCA studies, particularly those focused on technologies that could be deployed at scale. This proof-of-concept demonstrates how AI-assisted, evidence-grounded interpretation can support implementation-oriented decision-making beyond conventional LCA studies.

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.

Referee Report

2 major / 1 minor

Summary. The paper presents LCAi, a framework for AI-assisted life cycle assessment interpretation using perspective-conditioned retrieval-augmented generation. It fuses data from academic, industry, public discourse, and EU funding sources through a three-step process: (1) scenario anchor for system boundaries and targets, (2) perspective-specific micro-queries with constrained retrieval, and (3) neutral synthesis integrating only ledger-stored outputs. The framework is demonstrated on a hydrogen-enabled diesel reduction use case in an Italian apple production facility using GPT-5 nano, with the claim that it mitigates hallucination risk while preserving cross-domain diversity.

Significance. If empirically validated, the multi-perspective RAG architecture could address a real gap in LCA by providing structured, evidence-grounded translation of quantified hotspots into strategic pathways under uncertainty, extending AI tools beyond conventional quantification in environmental assessments.

major comments (2)
  1. [Abstract] Abstract: the claim that 'the structured retrieval and constrained synthesis are designed to mitigate the risk of hallucination while preserving cross-domain diversity' lacks any supporting quantitative validation, baseline comparisons, error metrics, or sensitivity analysis on the four datasets; the demonstration is described only as a single qualitative case study.
  2. [Methods (three-step process)] Three-step process description: the neutral synthesis step is asserted to integrate 'only ledger-stored outputs without further retrieval' to avoid new errors, but no details are supplied on ledger construction, output validation, or how perspective-specific results are weighted or checked for completeness, leaving the bias-mitigation claim untested.
minor comments (1)
  1. [Abstract] Abstract: 'GPT-5 nano' is referenced without clarification of its status, training cutoff, or relation to publicly available models.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below, indicating planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'the structured retrieval and constrained synthesis are designed to mitigate the risk of hallucination while preserving cross-domain diversity' lacks any supporting quantitative validation, baseline comparisons, error metrics, or sensitivity analysis on the four datasets; the demonstration is described only as a single qualitative case study.

    Authors: We agree that the current demonstration consists of a single qualitative case study and does not include quantitative validation, baselines, or error metrics. The manuscript is framed as a proof-of-concept for the perspective-conditioned RAG architecture rather than an empirical benchmarking study. The claim in the abstract refers to the design rationale of the three-step process. We will revise the abstract to clarify the illustrative nature of the demonstration and add a dedicated limitations subsection discussing the absence of quantitative evaluation and the scope for future work on metrics and sensitivity analysis across the datasets. revision: yes

  2. Referee: [Methods (three-step process)] Three-step process description: the neutral synthesis step is asserted to integrate 'only ledger-stored outputs without further retrieval' to avoid new errors, but no details are supplied on ledger construction, output validation, or how perspective-specific results are weighted or checked for completeness, leaving the bias-mitigation claim untested.

    Authors: The manuscript describes the neutral synthesis step at a high level to emphasize avoidance of additional retrieval. We acknowledge that explicit details on ledger construction, validation procedures, weighting of perspective outputs, and completeness checks are not provided. We will expand the Methods section with implementation specifics from the demonstration, including how ledger entries are generated and stored from the micro-query outputs, any cross-checks performed, and the rationale for treating the ledger as the sole input to synthesis. This will allow readers to assess the bias-mitigation approach more directly. revision: yes

Circularity Check

0 steps flagged

No derivation chain; architecture proposal has no equations or fitted quantities

full rationale

The paper describes a three-step perspective-conditioned RAG framework (scenario anchor, perspective-specific micro-queries with constrained retrieval, neutral ledger-only synthesis) for LCA interpretation and demonstrates it qualitatively on one Italian apple/hydrogen case. No equations, parameters, predictions, or first-principles derivations appear in the provided text. The contribution is an architectural proposal whose claims rest on design intent rather than any reduction of outputs to inputs by construction. No self-citations or uniqueness theorems are invoked as load-bearing elements. This is the normal non-circular outcome for a methods/architecture paper without a mathematical derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces a methodological architecture but does not specify numerical free parameters, new physical entities, or unproved mathematical axioms beyond standard assumptions about LLM behavior under retrieval constraints.

axioms (1)
  • domain assumption Large language models guided by constrained, perspective-specific retrieval and neutral synthesis can reduce hallucination while maintaining cross-domain diversity in domain-specific interpretation tasks.
    Invoked to justify the framework's design for mitigating hallucination risk in the interpretation phase.

pith-pipeline@v0.9.1-grok · 5798 in / 1322 out tokens · 26465 ms · 2026-06-26T05:10:13.650826+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 27 canonical work pages

  1. [3]

    V., Fleischmann, M., Wasserman, J., McBride, J., Gerard, J., \ & Leblanc, F

    Jordahl, K., den Bossche, J. V., Fleischmann, M., Wasserman, J., McBride, J., Gerard, J., \ & Leblanc, F. (2020). geopandas/geopandas: v0.8.1. Zenodo. https://doi.org/10.5281/zenodo.3946761

    work page doi:10.5281/zenodo.3946761 2020

  2. [5]

    Li, B., Jiang, Y., Gadepally, V., & Tiwari, D. (2024). LLM Inference Serving: Survey of Recent Advances and Opportunities. 2024 IEEE High Performance Extreme Computing Conference (HPEC), 1--8. https://doi.org/10.1109/HPEC62836.2024.10938426

    work page doi:10.1109/hpec62836.2024.10938426 2024

  3. [6]

    Neural Comput Appl 37:28191--28267

    Abo El-Enen M, Saad S, Nazmy T (2025) A survey on retrieval-augmentation generation (RAG) models for healthcare applications. Neural Comput Appl 37:28191--28267. https://doi.org/10.1007/s00521-025-11666-9

    work page doi:10.1007/s00521-025-11666-9 2025

  4. [7]

    J Clean Prod 229:886--901

    Algunaibet IM, Guill\' e n-Gos\' a lbez G (2019) Life cycle burden-shifting in energy systems designed to minimise greenhouse gas emissions. J Clean Prod 229:886--901. https://doi.org/10.1016/j.jclepro.2019.04.276

    work page doi:10.1016/j.jclepro.2019.04.276 2019

  5. [8]

    J Decis Syst 1--30

    Arslan M, Munawar S, Cruz C (2024) Business insights using RAG--LLMs: a review and case study. J Decis Syst 1--30. https://doi.org/10.1080/12460125.2024.2410040

    work page doi:10.1080/12460125.2024.2410040 2024

  6. [9]

    Chemical Engineering Journal 73:1--21

    Azapagic A (1999) Life cycle assessment and its application to process selection, design and optimisation. Chemical Engineering Journal 73:1--21. https://doi.org/10.1016/S1385-8947(99)00042-X

    work page doi:10.1016/s1385-8947(99)00042-x 1999

  7. [10]

    Balaguer A, Benara V, Cunha RL de F, et al (2024) RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture

    2024

  8. [11]

    In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics

    Bang Y, Ji Z, Schelten A, et al (2025) HalluLens: LLM Hallucination Benchmark. In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. pp 24128--24156

    2025

  9. [12]

    IEEE Trans Big Data 12:346--361

    Douze M, Guzhva A, Deng C, et al (2026) The Faiss Library. IEEE Trans Big Data 12:346--361. https://doi.org/10.1109/TBDATA.2025.3618474

    work page doi:10.1109/tbdata.2025.3618474 2026

  10. [13]

    https://doi.org/10.1016/J.PROCIR.2024.01.131

    Goridkov N, Wang Y, Goucher-Lambert K (2024) What's in this LCA Report? Procedia CIRP 122:964--969. https://doi.org/10.1016/J.PROCIR.2024.01.131

    work page doi:10.1016/j.procir.2024.01.131 2024

  11. [14]

    pp 143--158

    Hasan MdT, Waseem M, Kemell K-K, et al (2026) Engineering RAG Systems for Real-World Applications. pp 143--158

    2026

  12. [15]

    ISO 14044 (2014) Environmental management -- Life cycle assessment -- Requirement and guidelines

    2014

  13. [16]

    J MR, VM K, Warrier H, Gupta Y (2024) Fine Tuning LLM for Enterprise: Practical Guidelines and Recommendations

    2024

  14. [17]

    In: Findings of the Association for Computational Linguistics: EMNLP 2023

    Ji Z, Yu T, Xu Y, et al (2023) Towards Mitigating LLM Hallucination via Self Reflection. In: Findings of the Association for Computational Linguistics: EMNLP 2023. pp 1827--1843

    2023

  15. [18]

    J Clean Prod 1:143--149

    Keoleian GA (1993) The application of life cycle assessment to design. J Clean Prod 1:143--149. https://doi.org/10.1016/0959-6526(93)90004-U

    work page doi:10.1016/0959-6526(93)90004-u 1993

  16. [19]

    In: Streamlit for Web Development

    Khorasani M, Abdou M, Hern\' a ndez Fern\' a ndez J (2025) Streamlit Basics. In: Streamlit for Web Development. Apress, Berkeley, CA, pp 31--66

    2025

  17. [20]

    Kumar A, Kim S, Bakshi B. R. (2026). Role of artificial intelligence in the chemical industry transition to a sustainable, circular, and net-zero future. Current Opinion in Chemical Engineering, 51, 101234. https://doi.org/10.1016/j.coche.2026.101234

    work page doi:10.1016/j.coche.2026.101234 2026

  18. [21]

    Int J Life Cycle Assess 16:247--257

    Lewandowska A, Matuszak-Flejszman A, Joachimiak K, Ciroth A (2011) Environmental life cycle assessment as a tool for identification of environmental aspects in EMS. Int J Life Cycle Assess 16:247--257. https://doi.org/10.1007/s11367-011-0252-3

    work page doi:10.1007/s11367-011-0252-3 2011

  19. [22]

    Lewis P, Perez E, Piktus A, et al (2021) Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

    2021

  20. [23]

    In: Proceedings of the 47th ACM SIGIR Conference

    Lin X, Wang W, Li Y, et al (2024) Data-efficient Fine-tuning for LLM-based Recommendation. In: Proceedings of the 47th ACM SIGIR Conference. ACM, New York, NY, USA, pp 365--374

    2024

  21. [24]

    Results in Engineering 29:109568

    Luan K, Jing Y (2026) AI-enabled integration of life cycle assessment in building retrofitting. Results in Engineering 29:109568. https://doi.org/10.1016/j.rineng.2026.109568

    work page doi:10.1016/j.rineng.2026.109568 2026

  22. [25]

    Sustainability 15:6326

    Mohammadi Kashka F, Tahmasebi Sarvestani Z, Pirdashti H, et al (2023) Sustainable Systems Engineering Using LCA: Application of AI for Predicting Agro-Environmental Footprint. Sustainability 15:6326. https://doi.org/10.3390/su15076326

    work page doi:10.3390/su15076326 2023

  23. [26]

    Eur J Hortic Sci 87:1--22

    Muder A, Garming H, Dreisiebner-Lanz S, et al (2022) Apple production and apple value chains in Europe. Eur J Hortic Sci 87:1--22. https://doi.org/10.17660/eJHS.2022/059

    work page doi:10.17660/ejhs.2022/059 2022

  24. [27]

    J Open Source Softw 2:236

    Mutel C (2017) Brightway: An open source framework for Life Cycle Assessment. J Open Source Softw 2:236. https://doi.org/10.21105/joss.00236

    work page doi:10.21105/joss.00236 2017

  25. [28]

    AI 6:226

    Neha F, Bhati D, Shukla DK (2025) Retrieval-Augmented Generation (RAG) in Healthcare: A Comprehensive Review. AI 6:226. https://doi.org/10.3390/ai6090226

    work page doi:10.3390/ai6090226 2025

  26. [29]

    Journal of Sustainable Metallurgy 11:3590--3605

    Nwagwu CC, Ogorodnyk O, S lvsberg E, et al (2025) Integrating Artificial Intelligence into Life Cycle Assessment. Journal of Sustainable Metallurgy 11:3590--3605. https://doi.org/10.1007/s40831-025-01305-x

    work page doi:10.1007/s40831-025-01305-x 2025

  27. [30]

    Discover Sustainability 6:1237

    Petrosa D, Haverkamp P, Backes JG, et al (2025) Development of a BIM-based AI-driven matching tool for LCA datasets. Discover Sustainability 6:1237. https://doi.org/10.1007/s43621-025-02203-8

    work page doi:10.1007/s43621-025-02203-8 2025

  28. [31]

    Int J Life Cycle Assess 27:1164--1169

    Prado V, Seager TP, Guglielmi G (2022) Challenges and risks when communicating comparative LCA results to management. Int J Life Cycle Assess 27:1164--1169. https://doi.org/10.1007/s11367-022-02090-5

    work page doi:10.1007/s11367-022-02090-5 2022

  29. [32]

    S., You F

    Preuss N, Alshehri A. S., You F. (2024). Large language models for life cycle assessments: Opportunities, challenges, and risks. Journal of Cleaner Production, 466, 142824. https://doi.org/10.1016/j.jclepro.2024.142824

    work page doi:10.1016/j.jclepro.2024.142824 2024

  30. [33]

    Environ Sci Technol 60:33--48

    Preuss N, You F (2026) Automating Life Cycle Assessments through Artificial Intelligence Agents. Environ Sci Technol 60:33--48. https://doi.org/10.1021/acs.est.5c14493

    work page doi:10.1021/acs.est.5c14493 2026

  31. [34]

    In: Proceedings of EMNLP 2023

    Rawte V, Chakraborty S, Pathak A, et al (2023) The Troubling Emergence of Hallucination in Large Language Models. In: Proceedings of EMNLP 2023. pp 2541--2573

    2023

  32. [35]

    Ren Y, Sutherland DJ (2025) Learning Dynamics of LLM Finetuning

    2025

  33. [36]

    Environ Manage 29:132--142

    Ross S, Evans D (2002) Use of Life Cycle Assessment in Environmental Management. Environ Manage 29:132--142. https://doi.org/10.1007/s00267-001-0046-7

    work page doi:10.1007/s00267-001-0046-7 2002

  34. [37]

    In: The Illusion Engine

    S ekrst K (2025) Hallucinations. In: The Illusion Engine. Springer Nature Switzerland, Cham, pp 211--226

    2025

  35. [38]

    Int J Adv Manuf Technol

    Shafiq M, Ayub S, Muthevi A kumar, Prabhu MR (2024) AI-driven Life Cycle Assessment for sustainable hybrid manufacturing and remanufacturing. Int J Adv Manuf Technol. https://doi.org/10.1007/s00170-024-14930-9

    work page doi:10.1007/s00170-024-14930-9 2024

  36. [39]

    Applied Sciences 15:4234

    Swacha J, Gracel M (2025) Retrieval-Augmented Generation (RAG) Chatbots for Education: A Survey. Applied Sciences 15:4234. https://doi.org/10.3390/app15084234

    work page doi:10.3390/app15084234 2025

  37. [40]

    Environ Innov Soc Transit 45:154--169

    Verrier B, Li P-H, Pye S, Strachan N (2022) Incorporating social mechanisms in energy decarbonisation modelling. Environ Innov Soc Transit 45:154--169. https://doi.org/10.1016/j.eist.2022.10.003

    work page doi:10.1016/j.eist.2022.10.003 2022

  38. [41]

    Wernet, C

    Wernet G, Bauer C, Steubing B, et al (2016) The ecoinvent database version 3 (part I): overview and methodology. Int J Life Cycle Assess 21:1218--1230. https://doi.org/10.1007/s11367-016-1087-8

    work page doi:10.1007/s11367-016-1087-8 2016

  39. [42]

    pp 102--120

    Yu H, Gan A, Zhang K, et al (2025) Evaluation of Retrieval-Augmented Generation: A Survey. pp 102--120

    2025

  40. [43]

    Zhang J (2025) Small Language Models Offer Significant Potential for Science Community

    2025

  41. [44]

    J Clean Prod 529:146776

    Zhang X, Guo X, Zhao J, et al (2025) Intelligent application of large language model to life cycle assessment methodology. J Clean Prod 529:146776. https://doi.org/10.1016/J.JCLEPRO.2025.146776

    work page doi:10.1016/j.jclepro.2025.146776 2025