Pith. sign in

REVIEW 3 major objections 6 minor 91 references

EASI-RAG is a five-block agile method that lets small industrial firms deploy a RAG assistant in under a month, with no prior RAG experience on the team.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

EASI-RAG is a structured agile method for deploying RAG tools in industrial SMEs, validated by one case study where a no-experience team built a working assistant in three weeks.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Useful method artifact, but the 86% accuracy claim is an in-sample fitted number and should not be quoted. the 3 major comments →

arxiv 2508.21024 v1 pith:NNFXQ5WH submitted 2025-08-28 cs.CL cs.IR

An Agile Method for Implementing Retrieval Augmented Generation Tools in Industrial SMEs

Mathieu Bourdin , Anas Neumann , Thomas Paviot , Robert Pellerin , Samir Lamouri This is my paper
classification cs.CL cs.IR
keywords Retrieval-Augmented GenerationRAGSMEagile methodindustrial NLPLLM deploymentmethod engineeringcase study
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

EASI-RAG is a step-by-step, agile method for bringing retrieval-augmented generation (RAG) assistants into small and medium industrial companies that have no natural-language-processing expertise. The paper claims that by specifying who does what at each stage—users, a data expert, a process owner, and developers—and by iterating quickly through design, evaluation, error correction, integration, and feedback, a competent IT team can deliver a reliable document-answering tool in weeks rather than months. The supporting case study at an environmental analysis laboratory is the evidence: two part-time developers with no prior RAG experience produced a working assistant in about three weeks and roughly 70 person-hours, improving from 17 correct answers on the first run to zero incorrect answers after six targeted corrections. If the method generalises, SMEs could adopt RAG without large AI teams or budgets, while keeping the underlying procedure documents as the source of truth.

Core claim

The central claim is that RAG deployment in an SME is not primarily a modelling problem but a process problem, and that a method with explicit roles and iterative correction loops can solve it. EASI-RAG organizes the work into five blocks—initial design, evaluation, error analysis and correction, integration, and feedback—and assigns each activity to defined roles. A diagnostic table links each observed failure (missing data, chunking artefacts, unknown vocabulary, hallucination, irrelevant or incoherent answers, bad formatting) to concrete fixes such as hybrid sparse-dense retrieval, structure-aware chunking, prompt grounding instructions, and query rewriting. In the reported case, six modi

What carries the argument

The load-bearing object is the EASI-RAG method itself: five sequential blocks with an outer feedback loop and an inner correction loop. Around it, the diagnostic table in Section 3 is the mechanism that turns a wrong answer into a targeted repair—it classifies the failure by pipeline stage (retrieval vs generation) and prescribes a matching technique. Four roles (User, Data Expert, Process Owner, Developer) carry the method's claim that RAG projects fail for organisational reasons as much as technical ones.

Load-bearing premise

The method's evaluation relies on a 'data expert' who knows the documents deeply enough to write the test questions and expected answers; if a company lacks such a person, the quality loop cannot be trusted.

What would settle it

Have a second SME with similar procedure documents but no long-tenured data expert run EASI-RAG; if the test set cannot be built or the error-correction loop produces unreliable gold answers, the method's main support collapses. On a smaller scale, re-run the same nine-document case with a different data expert and compare the test-set annotations.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • An SME can go from zero RAG experience to a deployed assistant in roughly three weeks for about 70 person-hours of effort.
  • Simple starting defaults—fixed-size chunks, dense embeddings, a small retrieved context—are enough to reach a useful baseline; advanced techniques are only added when the diagnostic table points to them.
  • Hybrid retrieval (dense plus sparse) and an explicit 'use only the provided context' prompt instruction were the corrections that mattered most in the case.
  • The 'Report Incorrect Answer' feedback loop turns user reports into data updates, so the tool improves the reliability of the underlying procedure documents, not just the answers.
  • The two-loop agile structure lets the team avoid unimplemented advanced features (multimodal RAG, reranking, semantic chunking) that would have added cost without improving this use case.

Where Pith is reading between the lines

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

  • The method's dependence on a data expert who can author gold-standard test answers means SMEs without such a person may need to budget for document-curation work before evaluation can start.
  • The same five-block loop could be applied to other document-grounded assistants, such as regulatory-compliance question answering or new-employee training, where a known set of authoritative documents exists.
  • Because the case used a proprietary API model on non-sensitive data, the cost estimate holds only when confidentiality allows cloud calls; firms with sensitive procedures would need local models, shifting the cost balance.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes EASI-RAG (Enterprise Application Support for Industrial RAG), a five-block agile method—initial design, evaluation, error analysis and correction, integration, feedback gathering—with explicit roles, activities, techniques, and outputs. The method is aimed at industrial SMEs that lack NLP expertise and is validated through a single case study in an environmental testing laboratory. A team with no prior RAG experience built a RAG-based assistant over about three weeks, using nine procedural documents (≈30k tokens). On a 50-query test set, the system improved from 17 correct/19 acceptable/14 incorrect in the initial evaluation to 44 correct/7 acceptable/0 incorrect after six corrective modifications. The paper claims fast implementation, high user adoption, accurate answers, and low cost, with approximately 70 person-hours and minimal LLM API expenditure.

Significance. If the method is taken as a structured, role-aware template for RAG deployment in SMEs, it addresses a genuine gap: prior industrial RAG papers mostly report specific technical solutions rather than a repeatable method. Strengths include the explicit role definitions, the diagnostic Table 4 linking error symptoms to corrective techniques, the cost and effort breakdown in Section 4.8, and the authors' acknowledgment of the single-case limitation in Section 5. However, the evidence supporting the central claim of 'accurate answers' is weak: the only quantitative results come from a self-authored, in-sample test set used to drive iterative corrections, with no held-out evaluation, no external baseline, and no statistical analysis. The method's usefulness is plausible, but the manuscript currently overstates what the case demonstrates.

major comments (3)
  1. [§4.3 and §4.5] The central accuracy claim—44 correct, 7 acceptable, 0 incorrect out of 50—is measured on the same test set that was used to identify and correct errors through six successive modifications. Because the system was iterated until the predefined targets were met, these final numbers are in-sample fitted outcomes, not an unbiased estimate of real-world performance. This is not a purely theoretical concern: Section 4.7 reports that two new real-user questions produced incorrect answers in the first week of deployment, and those questions were then added to the official test set. The paper should either provide a held-out evaluation (e.g., a second set of queries not used during correction loops) or explicitly reframe the reported numbers as development-set performance and temper the accuracy claims in Sections 4.8 and 5 accordingly.
  2. [§4.1 and §4.3] The evaluation's gold standard depends on a single Data Expert—a senior team member with over 20 years of experience—who defines the expected answers and their locations in the documents. This same expert is also involved in the error-analysis loop. The method therefore implicitly assumes access to a high-availability, high-authority domain expert, and SMEs without such a person may not be able to reproduce the reported process. The paper does not discuss this dependence as a boundary condition. The authors should state this assumption explicitly and suggest how the method could be adapted when such an expert is unavailable.
  3. [§5 (and abstract/conclusion)] The authors acknowledge the single-case limitation, but the abstract, Section 4.8, and conclusion claim that EASI-RAG 'demonstrates' fast implementation, high user adoption, accurate answers, and enhanced data reliability. With one self-evaluated case and no comparison against any baseline—manual document search, a simpler retrieval method, or an off-the-shelf RAG tool—these claims exceed the evidence. The wording should be softened to 'suggests' or 'illustrates,' and the discussion should specify what additional evidence (e.g., multiple cases, external baselines, held-out evaluation) would be needed to establish generalizability.
minor comments (6)
  1. [Section numbering] Section 4.4 is missing; the section currently titled '4.5 Error analysis and correction' should be renumbered and cross-references updated.
  2. [Figure 3] The Pareto chart lacks axis labels and a clear definition of what the bars represent (number of errors vs. percentage of incorrect responses). Please clarify, especially since multiple error causes can contribute to a single incorrect answer.
  3. [References] Reference [21] and [41] are the same work (Wang et al., 2024, 'Searching for best practices in retrieval-augmented generation'). Merge and renumber.
  4. [§3, opening] The sentence 'According to Denner et al.'s definition of a method [28], if offers a systematic procedure' contains a typo: 'if' should be 'it'.
  5. [§4.8] The statement 'no unsatisfactory responses' is inconsistent with the definitions in Section 4.3, where 'acceptable' responses include partial answers or 'I don't know' statements. Seven acceptable responses are not correct; the sentence should be rephrased to avoid implying perfect performance.
  6. [§4.7] Adding the two new incorrect real-user questions to the 'official test set' further contaminates the evaluation set. If these questions are used in future iterations, the reported accuracy will become even less representative of steady-state performance. Clarify how the test set is kept static for valid comparisons.

Circularity Check

1 steps flagged

86% accuracy is an in-sample result: the same 50 queries drove the correction loop and then were used to report final performance.

specific steps
  1. fitted input called prediction [Section 4.3 and Section 4.5 (claimed in Section 4.8)]
    "The incorrect responses are subsequently analysed following the methodology outlined in Section 3. The analysis reveals ten distinct causes of error, which are summarised and illustrated in the following Figure 3. ... Identified errors have been corrected through six successive modifications, addressing the issues from the most frequent to the least frequent: ... These various modifications resulted in the following outcomes: • Average response time: approximately 2 seconds (unchanged); • 44 correct answers; • 7 acceptable answers; and • 0 incorrect answers. ... The deployed system achieved 86"

    The 50-query set defined in Section 4.3 is the same set whose failures are diagnosed and fixed in Section 4.5. The six modifications are targeted corrections to errors on those exact queries, and the final 44/7/0 results, plus the derived '86% correct' claim, are measured on that same set. No held-out or independently collected test set is used. The paper's own Section 4.7 shows the loop missed failure modes: two new real-user questions in the first week produced incorrect answers and were added to the official test set, so the headline numbers exclude known post-deployment failures. Thus the central accuracy claim is a fitted, in-sample outcome of the correction loop rather than an independent prediction of EASI-RAG's performance.

full rationale

The only load-bearing circularity is the evaluation protocol. The paper is otherwise a method-engineering contribution: EASI-RAG is defined from Zellner's method framework and a literature review, and its deployment-speed and cost claims (three weeks, ~70 person-hours) are not derived from the test set. The prior self-citation [5] is background and does not carry the numerical claim. The central accuracy validation, however, reduces by construction: the 50 queries defined in Section 4.3 are used in Section 4.5 for root-cause analysis; six corrective modifications are made to the pipeline on the basis of those queries; and the final 44/7/0 and '86% correct' are then reported on the same set. Because there is no held-out test set or externally falsifiable benchmark, the accuracy number is an in-sample artifact of the correction loop. Section 4.7 strengthens this reading: two new real-user failures were added to the official test set after deployment, showing that the 50-query set did not capture all failure modes and that the reported 'no unsatisfactory responses' excludes known failures. Score 6: one central 'prediction' (accuracy validation) reduces by construction; the rest of the method claim retains independent content.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The paper introduces an engineering method and role model rather than a new physical or mathematical entity. The relevant 'things pulled from upstream' are domain assumptions about SME resources and the role of a data expert, plus hand-chosen design parameters.

free parameters (4)
  • initial chunk size = 1000 tokens
    Hand-selected initial chunk size without overlap (Section 4.2), later replaced by hierarchical chunking. The reported accuracy is conditional on this choice.
  • number of retrieved chunks = 3 initially, 5 after correction
    Chosen to keep context within 2,000-3,000 tokens (Sections 4.2 and 4.5). The final result depends on this setting.
  • performance targets = >80% correct, <20% acceptable, 0 contradictions
    Arbitrary targets set by the project team in Section 4.3. The evaluation reports whether these self-defined targets are met.
  • embedding model = all-MiniLM-L6-v2
    Selected via MTEB and local testing (Section 4.2). Retrieval quality is tied to this model choice.
axioms (4)
  • domain assumption A data expert with deep document knowledge is available to author test queries and validate answers
    The evaluation and correction loops in Sections 4.3 and 4.5 depend on this role; Section 4.1 describes a 20-year expert, which may not be typical in SMEs.
  • domain assumption The 50 test queries authored by users are a valid proxy for real operational questions
    The accuracy claim rests entirely on this test set; no independent or longitudinal validation is given.
  • domain assumption Text-only RAG is sufficient for the target use cases
    The method scopes out images and temporal or numerical data (Section 1); the case study found text sufficient, but this may not generalize.
  • domain assumption RAG is preferable to fine-tuning for SMEs due to resource constraints
    Stated in Section 1 as a premise; the paper does not compare against fine-tuning.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of An Agile Method for Implementing Retrieval Augmented Generation Tools in Industrial SMEs." pith.science (2026). https://pith.science/paper/NNFXQ5WH

@misc{pith2026250821024,
  author       = {Pith},
  title        = {Pith review of: An Agile Method for Implementing Retrieval Augmented Generation Tools in Industrial SMEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NNFXQ5WH}},
  note         = {Machine review of arXiv:2508.21024}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Retrieval-Augmented Generation (RAG) has emerged as a powerful solution to mitigate the limitations of Large Language Models (LLMs), such as hallucinations and outdated knowledge. However, deploying RAG-based tools in Small and Medium Enterprises (SMEs) remains a challenge due to their limited resources and lack of expertise in natural language processing (NLP). This paper introduces EASI-RAG, Enterprise Application Support for Industrial RAG, a structured, agile method designed to facilitate the deployment of RAG systems in industrial SME contexts. EASI-RAG is based on method engineering principles and comprises well-defined roles, activities, and techniques. The method was validated through a real-world case study in an environmental testing laboratory, where a RAG tool was implemented to answer operators queries using data extracted from operational procedures. The system was deployed in under a month by a team with no prior RAG experience and was later iteratively improved based on user feedback. Results demonstrate that EASI-RAG supports fast implementation, high user adoption, delivers accurate answers, and enhances the reliability of underlying data. This work highlights the potential of RAG deployment in industrial SMEs. Future works include the need for generalization across diverse use cases and further integration with fine-tuned models.

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

91 extracted references · 55 canonical work pages · 4 internal anchors

  1. [1]

    knowledge

    Introduction Since the release of OpenAI’s ChatGPT at the end of 2022, public and research interest in language models has surged dramatically. Within just a few years, numerous new language models have emerged, including Meta’s LLaMa, Google’s Gemini, OpenAI’s GPT-4, Anthropic’s Claude, Mistral AI’s Mixtral, and the DeepSeek LLM family. These tools are i...

    2022

  2. [2]

    goal -oriented systematic approach, which helps to resolve theoretical and practical tasks

    Related work Several studies explore the use of Retrieval -Augmented Generation (RAG) in industrial contexts. Chen et al. [9] introduce an interactive knowledge management (IIKM) system to assist technicians with technical repairs and internal policy inquiries. Chaudhary et al. [10] designed a Llama-based chatbot for Continuous Integration and Continuous ...

  3. [3]

    Related Works

    EASI-RAG method This section presents the newly proposed EASI-RAG approach: a method for implementing RAG -based tools in Industrial SMEs. Following Denner et al.'s definition of a method [28], if offers a systematic procedure, including guidelines and step-by-step strategies, to implement solution. The proposed method is divided into five blocks (see Fig...

    work page 2025

  4. [4]

    Using the following contextual elements: [list of retrieved chunks, separated by

    Industrial test case To validate the relevance of the presented method, the EASI-RAG method was applied to a real -world use case in an industrial company. 4.1 Company and data description The industrial partner is an environmental analysis laboratory in France employing around 200 people. The case study focuses on the development of a virtual assistant d...

    work page 2017

  5. [5]

    Discussion The method proposed in this paper has proven to be both efficient and accessible, enabling the rapid deployment of a RAG system in a real-world setting using a small team with no experience in RAG technologies. Its agile and pragmatic design allows for the quick development of a functional tool that delivers correct answers in over 85% of cases...

    work page

  6. [6]

    Conclusion The EASI-RAG method presented in this paper describes an agile approach focused on the rapid deployment of RAG tools within the context of industrial SMEs. This method outlines the various activities required to implement RAG tools, specifying for each acti vity the possible techniques, the personnel involved, and the expected deliverables, fol...

    work page

  7. [7]

    Bauer, M., van Dinther, C., & Kiefer, D. (2020). Machine learning in SME: an empirical study on enablers and success factors

    work page 2020

  8. [8]

    F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P

    Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P. (2023). Lost in the middle: How language models use long contexts. arXiv preprint arXiv:2307.03172

    Pith/arXiv arXiv 2023

  9. [9]

    Application of retrieval-augmented generation for interactive industrial knowledge management via a large language model ◐ ○ ○ ● ○ ● ● ○

  10. [10]

    Developing a Llama-Based Chatbot for CI/CD Question Answering: A Case Study at Ericsson ● ● ○ ● ○ ● ● ○

  11. [11]

    Empowering LLMs by hybrid retrieval-augmented generation for domain- centric Q&A in smart manufacturing ● ● ○ ● ○ ● ● ○

  12. [12]

    Can a troubleshooting AI assistant improve task performance in industrial contexts? ● ● ○ ● ○ ● ● ○

  13. [13]

    LLM-MANUF: An integrated framework of Fine-Tuning large language models for intelligent Decision-Making in manufacturing ● ● ○ ● ◐ ○ ● ○

  14. [14]

    Retrieval-Augmented Generation-aided causal identification of aviation accidents: A large language model methodology ● ● ○ ● ● ● ◐ ○

  15. [15]

    A knowledge-graph enhanced large language model-based fault diagnostic reasoning and maintenance decision support pipeline towards industry 5.0 ● ● ○ ● ○ ● ● ○

  16. [16]

    ● ◐ ○ ● ● ● ◐ ○

    RAGVA: Engineering retrieval augmented generation-based virtual assistants in practice. ● ◐ ○ ● ● ● ◐ ○

  17. [17]

    A Survey on Knowledge-Oriented Retrieval-Augmented Generation ● ● ○ ● ○ ● ○ ○

  18. [18]

    Retrieval-Augmented Generation for Large Language Models: A Survey ● ● ○ ● ○ ● ○ ○

    work page

  19. [19]

    Generative AI-enabled supply chain management: The critical role of coordination and dynamism ◐ ◐ ◐ ○ ○ ○ ● ●

    work page

  20. [20]

    Are both generative AI and ChatGPT game changers for 21st -Century operations and supply chain excellence? ◐ ◐ ◐ ○ ○ ○ ● ●

    work page

  21. [21]

    Searching for Best Practices in Retrieval-Augmented Generation ● ● ○ ● ○ ● ○ ○

    work page

  22. [22]

    A federated machine learning approach for order-level risk prediction in Supply Chain Financing ● ● ○ ● ○ ○ ● ●

    work page

  23. [23]

    Technology isn't enough for Industry 4.0: on SMEs and hindrances to digital transformation ○ ◐ ◐ ◐ ○ ○ ● ●

    work page

  24. [24]

    A dual evolutionary perspective on the Co-evolution of data-driven digital transformation and value proposition in manufacturing SMEs ○ ◐ ◐ ● ○ ○ ● ●

    work page

  25. [25]

    Adoption paths of digital transformation in manufacturing SME ○ ◐ ◐ ◐ ○ ○ ● ● Legend: ● = The paper fully addresses this aspect ◐ = The paper partially addresses this aspect ○ = The paper does not address this aspect at all Table 1: Structured evaluation of related works

    work page

  26. [26]

    Beutel, G., Geerits, E., & Kielstein, J. T. (2023). Artificial hallucination: GPT on LSD?. Critical Care, 27(1), 148

    work page 2023

  27. [27]

    & Fung, P

    Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., ... & Fung, P. (2023). Survey of hallucination in natural language generation. ACM computing surveys, 55(12), 1-38

    work page 2023

  28. [28]

    M., Alghisi, S., & Riccardi, G

    Mousavi, S. M., Alghisi, S., & Riccardi, G. (2024). DyKnow: dynamically verifying time -sensitive factual knowledge in LLMs. arXiv preprint arXiv:2404.08700

    Pith/arXiv arXiv 2024

  29. [29]

    & Kiela, D

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., ... & Kiela, D. (2020). Retrieval -augmented generation for knowledge -intensive nlp tasks. Advances in neural information processing systems, 33, 9459-9474

    work page 2020

  30. [30]

    Bourdin, M., Neumann, A., Paviot, T., Pellerin, R., & Lamouri, S. (2024). Exploring the applications of natural language processing and language models for production, planning, and control activities of SMEs in industry 4.0: a systematic literature review. Journal of Intelligent Manufacturing

    work page 2024

  31. [31]

    Moeuf, A., Tamayo, S., Lamouri, S., Pellerin, R., & Lelievre, A. (2016). Strengths and weaknesses of small and medium sized enterprises regarding the implementation of lean manufacturing. IFAC-PapersOnLine, 49(12), 71-76

    work page 2016

  32. [32]

    C., Pardeshi, M

    Chen, L. C., Pardeshi, M. S., Liao, Y. X., & Pai, K. C. (2025). Application of retrieval -augmented generation for interactive industrial knowledge management via a large language model. Computer Standards & Interfaces, 94, 103995

    work page 2025

  33. [33]

    L., Thomas, D., Nejati, S., & Sabetzadeh, M

    Chaudhary, D., Vadlamani, S. L., Thomas, D., Nejati, S., & Sabetzadeh, M. (2024, October). Developing a Llama -Based Chatbot for CI/CD Question Answering: A Case Study at Ericsson. In 2024 IEEE International Conference on Software Maintenance and Evolution (ICSME) (pp. 707-718). IEEE

    work page 2024

  34. [34]

    Wan, Y., Chen, Z., Liu, Y., Chen, C., & Packianather, M. (2025). Empowering LLMs by hybrid retrieval-augmented generation for domain-centric Q&A in smart manufacturing. Advanced Engineering Informatics, 65, 103212

    work page 2025

  35. [35]

    Löwhagen, N., Schwendener, P., & Netland, T. (2025). Can a troubleshooting AI assistant improve task performance in industria l contexts?. International Journal of Production Research, 1-22

    work page 2025

  36. [36]

    Du, K., Yang, B., Xie, K., Dong, N., Zhang, Z., Wang, S., & Mo, F. (2025). LLM-MANUF: An integrated framework of Fine-Tuning large language models for intelligent Decision-Making in manufacturing. Advanced Engineering Informatics, 65, 103263

    work page 2025

  37. [37]

    Ren, T., Zhang, Z., Jia, B., & Zhang, S. (2025). Retrieval-Augmented Generation-aided causal identification of aviation accidents: A large language model methodology. Expert Systems with Applications, 278, 127306

    work page 2025

  38. [38]

    Ma, Y., Zheng, S., Yang, Z., Pan, H., & Hong, J. (2025). A knowledge -graph enhanced large language model -based fault diagnostic reasoning and maintenance decision support pipeline towards industry 5.0. International Journal of Production Research, 1-22

    work page 2025

  39. [39]

    Yang, R., Fu, M., Tantithamthavorn, C., Arora, C., Vandenhurk, L., & Chua, J. (2025). RAGVA: Engineering retrieval augmented generation-based virtual assistants in practice. arXiv preprint arXiv:2502.14930

    Pith/arXiv arXiv 2025

  40. [40]

    & Chen, E

    Cheng, M., Luo, Y., Ouyang, J., Liu, Q., Liu, H., Li, L., ... & Chen, E. (2025). A survey on knowledge -oriented retrieval -augmented generation. arXiv preprint arXiv:2503.10677

    Pith/arXiv arXiv 2025

  41. [41]

    & Wang, H

    Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., ... & Wang, H. (2023). Retrieval -augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997, 2(1)

    Pith/arXiv arXiv 2023

  42. [42]

    E., & Zhang, Q

    Li, L., Liu, Y., Jin, Y., Cheng, T. E., & Zhang, Q. (2024). Generative AI -enabled supply chain management: The critical role of coordination and dynamism. International Journal of Production Economics, 277, 109388

    work page 2024

  43. [43]

    F., Queiroz, M

    Wamba, S. F., Queiroz, M. M., Jabbour, C. J. C., & Shi, C. V. (2023). Are both generative AI and ChatGPT game changers for 21st-Century operations and supply chain excellence?. International Journal of Production Economics, 265, 109015

    work page 2023

  44. [45]

    Kong, L., Zheng, G., & Brintrup, A. (2024). A federated machine learning approach for order -level risk prediction in supply chain financing. International Journal of Production Economics, 268, 109095

    work page 2024

  45. [46]

    K., Christiansen, L., & Lassen, A

    Hansen, A. K., Christiansen, L., & Lassen, A. H. (2024). Technology isn't enough for Industry 4.0: on SMEs and hindrances to digital transformation. International Journal of Production Research, 1-21

    work page 2024

  46. [47]

    Z., KAMAL, M

    ZHENG, J., ZHANG, J. Z., KAMAL, M. M., & MANGLA, S. K. (2025). A Dual Evolutionary Perspective on the Co -Evolution of Data -Driven Digital Transformation and Value Proposition in Manufacturing SMEs. International Journal of Production Economics, 109561

    work page 2025

  47. [48]

    Battistoni, E., Gitto, S., Murgia, G., & Campisi, D. (2023). Adoption paths of digital transformation in manufacturing SME. International Journal of Production Economics, 255, 108675

    work page 2023

  48. [49]

    ( 2005, March)

    Braun, C., Wortmann, F., Hafner, M., & Winter, R. ( 2005, March). Method construction -a core approach to organizational engineering. In Proceedings of the 2005 ACM symposium on Applied computing (pp. 1295-1299)

    work page 2005

  49. [50]

    Zellner, G. (2011). A structured evaluation of business process improvement approaches. Business process management journal, 17(2), 203-237

    work page 2011

  50. [51]

    S., Püschel, L

    Denner, M. S., Püschel, L. C., & Röglinger, M. (2018). How to exploit the digitalization potential of business processes. Business & Information Systems Engineering, 60, 331-349

    work page 2018

  51. [52]

    Manifesto, A. (2001). Manifesto for agile software development

    work page 2001

  52. [53]

    Barzi, R. (2011). PME et agilité organisationnelle: étude exploratoire. Innovations, 35(2), 29-45

    work page 2011

  53. [54]

    J., You, Y., Milczek, J., Laverde, S., & Li, R

    Yepes, A. J., You, Y., Milczek, J., Laverde, S., & Li, R. (2024). Financial report chunking for effective retrieval augmented generation. arXiv preprint arXiv:2402.05131

    Pith/arXiv arXiv 2024

  54. [55]

    Płonka, M., Kocot, K., Hołda, K., Daniec, K., & Nawrat, A. (2025). A comparative evaluation of the effectiveness of document splitters for large language models in legal contexts. Expert Systems with Applications, 126711

    work page 2025

  55. [56]

    A., Nossum, A

    Brådland, H., Goodwin, M., Andersen, P. A., Nossum, A. S., & Gupta, A. (2025, July). A New HOPE: Domain -agnostic Automatic Evaluation of Text Chunking. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 170-179)

    work page 2025

  56. [57]

    Ahmed, T., & Choudhury, S. (2025). CHORUS: Zero -shot Hierarchical Retrieval and Orchestration for Generating Linear Programming Code. arXiv preprint arXiv:2505.01485

    Pith/arXiv arXiv 2025

  57. [58]

    Qu, R., Tu, R., & Bao, F. (2024). Is semantic chunking worth the computational cost?. arXiv preprint arXiv:2410.13070

    Pith/arXiv arXiv 2024

  58. [59]

    R., Rudat, M., Spiekermann, J., & Flores -Herr, N

    Bhat, S. R., Rudat, M., Spiekermann, J., & Flores -Herr, N. (2025). Rethinking Chunk Size For Long -Document Retrieval: A Multi-Dataset Analysis. arXiv preprint arXiv:2505.21700

    Pith/arXiv arXiv 2025

  59. [60]

    A RAG-Based Institutional Assistant

    Kuratomi, G., Pirozelli, P., Cozman, F. G., & Peres, S. M. (2025). A RAG-Based Institutional Assistant. arXiv preprint arXiv:2501.13880

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  60. [61]

    Álvaro, J. A. H., & Barreda, J. G. (2025). An advanced retrieval -augmented generation system for manufacturing quality control. Advanced Engineering Informatics, 64, 103007

    work page 2025

  61. [62]

    G., Zerhoudi, S., Mitrovic, J., & Granitzer, M

    Caspari, L., Dastidar, K. G., Zerhoudi, S., Mitrovic, J., & Granitzer, M. (2024). Beyond benchmarks: Evaluating embedding model similarity for retrieval augmented generation systems. arXiv preprint arXiv:2407.08275

    Pith/arXiv arXiv 2024

  62. [63]

    Muennighoff, N., Tazi, N., Magne, L., & Reimers, N. (2022). MTEB: Massive text embedding benchmark. arXiv preprint arXiv:2210.07316

    Pith/arXiv arXiv 2022

  63. [64]

    & Huang, X

    Wang, X., Wang, Z., Gao, X., Zhang, F., Wu, Y., Xu, Z., ... & Huang, X. (2024). Searching for best practices in retrieval -augmented generation. arXiv preprint arXiv:2407.01219

    Pith/arXiv arXiv 2024

  64. [65]

    & Schmidt, D

    White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert, H., ... & Schmidt, D. C. (2023). A prompt pattern catalog to en hance prompt engineering with chatgpt. arXiv preprint arXiv:2302.11382

    Pith/arXiv arXiv 2023

  65. [66]

    (2023, June)

    Marvin, G., Hellen, N., Jjingo, D., & Nakatumba -Nabende, J. (2023, June). Prompt engineering in large language models. In International conference on data intelligence and cognitive informatics (pp. 387-402). Singapore: Springer Nature Singapore

    work page 2023

  66. [67]

    Park, J., Atarashi, K., Takeuchi, K., & Kashima, H. (2025). Emulating Retrieval Augmented Generation via Prompt Engineering for Enhanced Long Context Comprehension in LLMs. arXiv preprint arXiv:2502.12462

    Pith/arXiv arXiv 2025

  67. [68]

    Chen, K., Zhou, X., Lin, Y., Feng, S., Shen, L., & Wu, P. (2025). A Survey on Privacy Risks and Protection in Large Language Models. arXiv preprint arXiv:2505.01976

    Pith/arXiv arXiv 2025

  68. [69]

    Torres, J. J. G., Bîndilă, M. B., Hofstee, S., Szondy, D., Nguyen, Q. H., Wang, S., & Englebienne, G. (2024). Automated Quest ion-Answer Generation for Evaluating RAG -based Chatbots. In 1st Workshop on Patient -Oriented Language Processing, CL4Health 2024 (pp. 204 -214). European Language Resources Association (ELRA)

    work page 2024

  69. [70]

    & Sun, M

    Yu, S., Tang, C., Xu, B., Cui, J., Ran, J., Yan, Y., ... & Sun, M. (2024). Visrag: Vision -based retrieval -augmented generation on multi -modality documents. arXiv preprint arXiv:2410.10594

    Pith/arXiv arXiv 2024

  70. [71]

    Yu, X., Jian, P., & Chen, C. (2025). TableRAG: A Retrieval Augmented Generation Framework for Heterogeneous Document Reasonin g. arXiv preprint arXiv:2506.10380

    arXiv 2025

  71. [72]

    Zhao, J., Ji, Z., Feng, Y., Qi, P., Niu, S., Tang, B., ... & Li, Z. (2024). Meta -chunking: Learning efficient text segmentation via logical perception. arXiv preprint arXiv:2410.12788

    Pith/arXiv arXiv 2024

  72. [73]

    Verma, P. (2025). S2 Chunking: A Hybrid Framework for Document Segmentation Through Integrated Spatial and Semantic Analysis. arXiv preprint arXiv:2501.05485

    Pith/arXiv arXiv 2025

  73. [74]

    Zhao, J., Ji, Z., Fan, Z., Wang, H., Niu, S., Tang, B., ... & Li, Z. (2025). MoC: Mixtures of Text Chunking Learners for Retr ieval-Augmented Generation System. arXiv preprint arXiv:2503.09600

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  74. [75]

    Ram, O., Levine, Y., Dalmedigos, I., Muhlgay, D., Shashua, A., Leyton -Brown, K., & Shoham, Y. (2023). In -context retrieval -augmented language models. Transactions of the Association for Computational Linguistics , 11, 1316-1331

    work page 2023

  75. [76]

    (2023, July)

    Gao, L., Ma, X., Lin, J., & Callan, J. (2023, July). Precise zero -shot dense retrieval without relevance labels. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 1762-1777)

    work page 2023

  76. [77]

    & Cheng, J

    Huang, H., Huang, Y., Yang, J., Pan, Z., Chen, Y., Ma, K., ... & Cheng, J. (2025). Retrieval -Augmented Generation with Hierarchical Knowledge. arXiv preprint arXiv:2503.10150

    arXiv 2025

  77. [78]

    Golden-Retriever: High-Fidelity Agentic Retrieval Augmented Generation for Industrial Knowledge Base

    An, Z., Ding, X., Fu, Y. C., Chu, C. C., Li, Y., & Du, W. (2024). Golden -Retriever: High-Fidelity Agentic Retrieval Augmented Generation for Industrial Knowledge Base. arXiv preprint arXiv:2408.00798

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  78. [79]

    Łajewska, W., & Balog, K. (2025). Ginger: Grounded information nugget -based generation of responses. arXiv preprint arXiv:2503.18174

    Pith/arXiv arXiv 2025

  79. [80]

    Tran, H., Wang, J., Ting, Y., Huang, W., & Chen, T. (2024). LEAF: Learning and Evaluation Augmented by Fact-Checking to Improve Factualness in Large Language Models. arXiv preprint arXiv:2410.23526

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  80. [81]

    (2023, December)

    Ma, X., Gong, Y., He, P., Zhao, H., & Duan, N. (2023, December). Query rewriting in retrieval -augmented large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (pp. 5303-5315)

    work page 2023

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.