Pith. sign in

REVIEW 4 major objections 5 minor 89 references

DBMS-LLM Integration Strategies in Industrial and Business Applications: Current Status and Future Challenges

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper argues that DBMS-LLM integration in industry falls into five architectural patterns, each with distinct coupling and trade-offs.

desk verdict Useful survey of DBMS-LLM integration, but the five-way taxonomy is a list of overlapping perspectives, not a principled partition; still deserves a serious referee. read the letter →

arxiv 2507.19254 v1 pith:Y22VT35A submitted 2025-07-25 cs.DB

Zhengtong Yan , Gongsheng Yuan , Qingsong Guo , Jiaheng Lu This is my paper · ORCID
classification cs.DB
keywords DBMS-LLMintegrationarchitecturalpatternsdataandAIText-to-SQLretrieval-augmentedgenerationqueryoptimizationmiddlewareorchestrationcloudplatformservices
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the many ways of connecting database management systems (DBMSs) and large language models (LLMs) in industry can be grouped into five representative architectural patterns: DB-first, LLM-first, middle-layer, pipe-connected, and platform-based. The driving claim is that this five-way split captures the real design space, covering where integration happens, how tightly the two systems are coupled, and what trade-offs result. A sympathetic reader would care because the choice among these patterns determines latency, security, maintainability, and how easily a system can evolve, and the paper turns that choice into a structured comparison rather than a set of anecdotes. The paper also derives open challenges from each pattern, such as cost modeling for LLM-backed operators and reliability of LLM tool use.

What carries the argument

The organizing device is a five-category taxonomy built on three dimensions: the purpose of integration, the system layer where integration occurs, and the degree of coupling. The taxonomy is supported by a comparison table of strengths and weaknesses and a feature-scoring table that rates each pattern as strong, moderate, or weak on coupling, real-time capability, scalability, extensibility, complexity, and security. These tables do the argument's work: they convert a set of system examples into a decision framework and ground the paper's later identification of pattern-specific open challenges.

What would settle it

Catalog every DBMS-LLM integration described in a broad corpus, such as all systems mentioned in recent published surveys, and check each against the five patterns; if a substantial number of working systems fit no pattern without forcing, or if two of the five patterns consistently appear together in practice, the taxonomy's claim to represent the landscape fails.

Watch

Extended reading notes

Core claim

The central claim is that DBMS-LLM integration in industrial and business settings is not a single technique but a landscape of five architectural patterns, each with a distinct structural relationship between the two systems. In DB-first integration the LLM is embedded inside the database as an interface, optimizer, UDF, or native operator; in LLM-first integration the database acts as a retrieval or cache backend for an LLM-driven application; middle-layer integration inserts an orchestrator between them; pipe-connected integration links them as independent services through data pipelines; and platform-based integration provides both as managed services on one cloud platform. The paper further claims that these patterns differ systematically along dimensions such as coupling degree, real-time performance, scalability, extensibility, complexity, and security, and that no single pattern dominates: the right choice depends on task requirements, functional needs, LLM dependencies, and deployment mode. A corollary of the taxonomy is that hybrid combinations are common and that the categories are explicitly non-orthogonal.

Load-bearing premise

The five-way taxonomy is assumed to be complete and representative, but the paper never derives it from an exhaustive survey or a formal design space; if a major integration mode is missing or the categories overlap too much to be useful, the main contribution collapses.

Editorial extensions

If this is right

  • DB-first integration, with LLMs as operators or UDFs, can extend SQL to semantic and multimodal operations while keeping database governance and indexing benefits.
  • LLM-first integration suits conversational and retrieval-augmented generation applications but sacrifices fine-grained query control and transactional guarantees.
  • Middle-layer integration makes pipelines composable and modular at the cost of added complexity, latency, and debugging difficulty.
  • Pipe-connected integration fits event-driven, high-throughput scenarios but accepts higher latency and harder failure recovery.
  • Platform-based integration lowers setup cost and offers scalability but risks vendor lock-in and limited transparency.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension implied by the paper's non-orthogonality admission is that hybrid patterns, not pure ones, may dominate real deployments, so future work could quantify how often production systems mix two or more of the five patterns.
  • The taxonomy could be operationalized as a selection agent: given workload latency, data volume, security needs, and user expertise, it could recommend one pattern; the paper mentions this as future work but does not specify the decision procedure.
  • The paper's cost-modeling challenge suggests a measurable benchmark: compare end-to-end query latency and accuracy of DB-first versus middle-layer integration on the same hybrid workload, something no standardized benchmark currently exists for.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper surveys recent work on integrating Database Management Systems (DBMSs) and Large Language Models (LLMs) in industrial and business applications. Its main contribution is a taxonomy of five architectural patterns — DB-first, LLM-first, Middle-layer, Pipe-connected, and Platform-based — along with qualitative comparisons, selection guidance, and a discussion of open challenges. The paper makes no empirical claims; the value rests on the coherence and usefulness of the proposed taxonomy and on the accuracy of the survey coverage.

Significance. If the proposed five-way taxonomy were convincing, the paper would provide a useful organizing framework for a rapidly growing area and could help practitioners choose among integration strategies. The survey covers a broad set of recent systems and honestly acknowledges that the categories are non-orthogonal and that hybrid combinations are common. The comparison tables and the enumerated future challenges are useful entry points. However, the taxonomy is the central load-bearing claim, and as detailed in the major comments it currently mixes heterogeneous classification criteria and contains internal inconsistencies. The contribution is therefore promising but not yet demonstrated in the present form.

major comments (4)
  1. [Section 3.1, Figure 1, Table 2] The five categories are not derived systematically from the three dimensions stated in Section 3.1 (purposes, system layer, coupling degree). The categories mix at least four distinct criteria: coupling (DB-first vs. LLM-first), presence of an orchestration layer (Middle-layer), dataflow topology (Pipe-connected), and deployment environment (Platform-based). The manuscript itself concedes the categorization is non-orthogonal. As a result, the central claim that these are 'five representative architectural patterns' is not established; the paper currently presents overlapping perspectives rather than a partition of a design space. Please either derive the categories from a single coherent set of dimensions or explicitly reframe them as orthogonal concerns rather than mutually exclusive patterns.
  2. [Section 3.6 vs. Section 3.2] The Platform-based category is not an architectural pattern in the same sense as the other four; it denotes managed cloud offerings (e.g., Snowflake Cortex, Google Cloud, Oracle Cloud) that can host any of the other patterns. For example, an LLM-as-UDF system described in Section 3.2 would belong to both DB-first and Platform-based if it were deployed on a cloud platform such as Snowflake Cortex. This conflation weakens the taxonomy and the selection guidance in Section 3.7. Please clarify whether Platform-based is a deployment axis orthogonal to the other four and adjust the claims and Table 2 accordingly.
  3. [Section 3.7 vs. Table 3] The selection guidance contradicts the comparison table. Section 3.7 states that 'Real-time and low-latency tasks are better served by DB-first, pip-connected, or platform-based integrated solutions,' but Table 3 rates Pipe-connected as ≈ (Moderate) on Real-Time, and Table 2 lists 'High latency or eventual consistency' as a weakness of Pipe-connected. This inconsistency is directly relevant to the paper's practical guidance and should be reconciled.
  4. [Table 3] The qualitative Strong/Moderate/Weak scores in Table 3 are presented without a methodology or supporting citations. The 'Complexity' row is especially ambiguous: for most rows a Strong score is desirable, but for Complexity a Strong score would presumably be undesirable, yet the symbol legend defines ✓/≈/× only as Strong/Moderate/Weak. Since the paper's practical contribution includes choosing integration strategies (Section 3.7), these unsupported and sign-ambiguous scores weaken the advice and need justification or a clear convention.
minor comments (5)
  1. [Throughout references] Several citations use given names instead of surnames, e.g., 'Kaushikpresent et al. [53]' (should be Rajan et al.), 'Moreh et al. [48]' (should be Park et al.), 'Jinyang et al. [35]' (should be Li et al.), 'Zhaodonghui et al. [36]' (Li et al.), 'Wenbo et al. [60]' (Sun et al.), 'Alekh et al. [31]' (Jindal et al.), 'Xinyang et al. [79]' (Zhao et al.), 'Sumedh et al. [54]' (Rasal et al.), 'Jiayi et al. [72]' (Wang et al.), 'Kai et al. [71]' (Waehner et al.), and 'Simone et al. [46]' (Papicchio et al.). Please standardize to surname-based citations.
  2. [Equation (1)] Equation (1), 'LLMs + DBMSs → Dual Infrastructures of Enterprises,' is not a mathematical equation and adds no information beyond the prose; consider removing it or rewriting the statement in words.
  3. [Figure 1] Figure 1 is very dense, and some labels such as 'DB-side fusion' do not match the terminology used elsewhere ('DB-first'). The arrows for data flow versus control flow are hard to distinguish at print size. Please enlarge the figure, align labels with the text, and make the legend more readable.
  4. [Section 2] The related work section lists prior surveys but does not clearly differentiate the proposed five-category taxonomy from the three paradigms of [82] or the interaction paradigms of [34]. A short positioning paragraph or a comparison table would strengthen the novelty claim.
  5. [Section 3.6 and Section 3.7] There is a typo in Section 3.6: 'This create significant opportunities' should be 'This creates significant opportunities.' Also, 'pip-connected' appears in Section 3.7, which is inconsistent with 'Pipe-connected' used elsewhere.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the five-pattern taxonomy is a survey classification built from cited systems and qualitative trade-offs, not a derivation that reduces to its own inputs.

full rationale

This paper is a survey, not a derivation or prediction pipeline. Its central claim is a taxonomy of five DBMS-LLM integration strategies, introduced in Section 3.1 as follows: 'Based on these dimensions, we identify five major integration strategies, as illustrated in Figure 1.' The categories are then populated with examples from the surveyed literature, and the paper explicitly notes that 'our categorization is non-orthogonal' and that hybrid combinations occur. That is standard survey classification practice, not circular reasoning. No equation is derived from an input and then relabeled as a prediction; no fitted parameter is used as an output; and no uniqueness theorem or prior result is invoked to force the taxonomy. The only author self-citation is [75] in the Related Work section ('Yan et al. [75] focus specifically on DRL-based techniques for join order selection, a key challenge in query optimization'), which is background context and is not load-bearing for the five-pattern claim. Concerns about whether the five categories are complete, mutually consistent, or derived cleanly from the stated dimensions are validity and framing concerns, not circularity; they do not amount to a reduction of the paper's conclusions to its assumptions. Accordingly, no circular step can be exhibited, and the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no free parameters, no postulates, and no new entities. The central claim is a taxonomy, and the taxonomy's categories are the paper's main 'construction'. The only implicit assumptions are that the five categories are well chosen and that the qualitative comparisons are accurate, both of which are reasonable but unverified premises for a survey.

assumptions (3)
  • ad hoc to paper The five proposed categories (DB-first, LLM-first, middle-layer, pipe-connected, platform-based) are assumed to be representative and useful.
    Introduced in Section 3.1 without a systematic derivation from a design space or a survey protocol. The entire paper is structured around this assumption.
  • domain assumption Qualitative ratings in Table 3 (Strong/Moderate/Weak) are assumed accurate.
    The ratings for coupling, real-time, scalability, extensibility, complexity, and security rest on the authors' interpretation of each pattern and are not backed by citations or controlled measurements.
  • domain assumption The cited papers accurately represent the state of the art in DBMS-LLM integration.
    A survey relies on the accuracy of the primary sources it summarises; the paper does not re-verify results from the cited systems.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DBMS-LLM Integration Strategies in Industrial and Business Applications: Current Status and Future Challenges." pith.science (2026). https://pith.science/paper/Y22VT35A

@misc{pith2026250719254,
  author       = {Pith},
  title        = {Pith review of: DBMS-LLM Integration Strategies in Industrial and Business Applications: Current Status and Future Challenges},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y22VT35A}},
  note         = {Machine review of arXiv:2507.19254}
}
read the original abstract

Modern enterprises are increasingly driven by the DATA+AI paradigm, in which Database Management Systems (DBMSs) and Large Language Models (LLMs) have become two foundational infrastructures powering a wide range of industrial and business applications, such as enterprise analytics, intelligent customer service, and data-driven decision-making. The efficient integration of DBMSs and LLMs within a unified system offers significant opportunities but also introduces new technical challenges. This paper surveys recent developments in DBMS+LLM integration and identifies key future challenges. Specifically, we categorize five representative architectural patterns based on their core design principles, strengths, and trade-offs. Based on this analysis, we further highlight several critical open challenges. We aim to provide a systematic understanding of the current integration landscape and to outline the unresolved issues that must be addressed to achieve scalable and efficient integration of traditional data management and advanced language reasoning in future intelligent applications.

Figures

Figures reproduced from arXiv: 2507.19254 by the authors.

Figure 1
Figure 1. Illustrations of different integration strategies. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The architecture and processing pipelines of DBMSs, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

89 extracted references · 44 canonical work pages

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. GPT-4 Technical Report. arXiv preprint arXiv:2303.08774 (2023). https://doi.org/10.48550/arXiv.2303.08774

    arXiv doi:10.48550/arxiv.2303.08774 2023

  2. [2]

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katherine Millican, Malcolm Reynolds, et al

  3. [3]

    Amazon Aurora. [n.d.]. https://aws.amazon.com/cn/rds/aurora/

  4. [4]

    Mouna Ammar, Christopher Rost, Riccardo Tommasini, Shubhangi Agarwal, Angela Bonifati, Petra Selmer, and Erhard Rahm. 2025. Towards Hybrid Graphs: Unifying Property Graphs and Time Series. In 28th International Conference on Extending Database Technology (EDBT) . 2483–2490. https://openproceedings. org/2025/conf/edbt/paper-183.pdf

    2025

  5. [5]

    Anthropic: Claude. [n.d.]. https://www.anthropic.com/

  6. [6]

    Apache Flink. [n.d.]. https://flink.apache.org/

  7. [7]

    Apache Kafka. [n.d.]. https://kafka.apache.org/

  8. [8]

    Apache Spark. [n.d.]. https://spark.apache.org/

Show all 89 references
  1. [9]

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen Technical Report. arXiv preprint arXiv:2309.16609 (2023)

    2023 arXiv

  2. [10]

    BigQuery. [n.d.]. https://cloud.google.com/bigquery

  3. [11]

    Asim Biswal, Liana Patel, Siddarth Jha, Amog Kamsetty, Shu Liu, Joseph E Gonzalez, Carlos Guestrin, and Matei Zaharia. 2025. Text2SQL is Not Enough: Unifying AI and Databases with TAG. In CIDR

    2025

  4. [12]

    Qingpeng Cai, Can Cui, Yiyuan Xiong, Wei Wang, Zhongle Xie, and Meihui Zhang. 2022. A Survey on Deep Reinforcement Learning for Data Processing and Analytics. IEEE Transactions on Knowledge and Data Engineering 35, 5 (2022), 4446–4465

    2022

  5. [13]

    Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gau- rav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sut- ton, Sebastian Gehrmann, et al . 2023. PaLM: Scaling Language Modeling with Pathways. Journal of Machine Learning Research 24, 240 (2023),...

    2023

  6. [14]

    ClickHouse. [n.d.]. https://clickhouse.com/

  7. [15]

    Edgar F Codd. 1970. A Relational Model of Data Large Shared Data Banks. Commun. ACM 13, 6 (1970), 377–387

    1970

  8. [16]

    Haowen Dong, Chao Zhang, Guoliang Li, and Huanchen Zhang. 2024. Cloud- Native Databases: A Survey. IEEE Transactions on Knowledge and Data Engineer- ing 36, 12 (2024), 7772–7791. https://doi.org/10.1109/TKDE.2024.3397508

    2024

  9. [17]

    Anas Dorbani, Sunny Yasser, Jimmy Lin, and Amine Mhedhbi. 2025. Beyond Quacking: Deep Integration of Language Models and RAG into DuckDB. arXiv preprint arXiv:2504.01157 (2025)

    2025 arXiv

  10. [18]

    DuckDB. [n.d.]. https://duckdb.org/

  11. [19]

    Luciano Floridi and Massimo Chiriatti. 2020. GPT-3: Its Nature, Scope, Limits, and Consequences. Minds and Machines 30, 4 (2020), 681–694

    2020

  12. [20]

    Yannis Foufoulas and Alkis Simitsis. 2023. Efficient Execution of User-Defined Functions in SQL Queries. Proceedings of the VLDB Endowment 16, 12 (2023), 3874–3877

    2023

  13. [21]

    Kai Franz, Samuel Arch, Denis Hirn, Torsten Grust, Todd C Mowry, and Andrew Pavlo. 2024. Dear User-Defined Functions, Inlining isn’t working out so great for us. Let’s try batching to make our relationship work. Sincerely, SQL. In Conference on Innovative Data Systems Research (CIDR)

    2024

  14. [22]

    Parker Glenn, Parag Dakle, Liang Wang, and Preethi Raghavan. 2024. BlendSQL: A Scalable Dialect for Unifying Hybrid Question Answering in Relational Algebra. In Findings of the Association for Computational Linguistics ACL 2024 . 453–466

    2024

  15. [23]

    Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. 2024. ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools. arXiv preprint arXiv:2406.12793 (2024)

    2024 arXiv

  16. [24]

    Yu Gu, Yiheng Shu, Hao Yu, Xiao Liu, Yuxiao Dong, Jie Tang, Jayanth Srinivasa, Hugo Latapie, and Yu Su. 2024. Middleware for LLMs: Tools Are Instrumental for Language Agents in Complex Environments. arXiv preprint arXiv:2402.14672 (2024)

    2024 arXiv

  17. [25]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al . 2025. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv preprint arXiv:2501.12948 (2025)

    2025 arXiv

  18. [26]

    Yikun Han, Chunjiang Liu, and Pengfei Wang. 2023. A Comprehensive Survey on Vector Database: Storage and Retrieval Technique, Challenge. arXiv preprint arXiv:2310.11703 (2023)

    2023

  19. [27]

    Mohamed S Hassan, Tatiana Kuznetsova, Hyun Chai Jeong, Walid G Aref, and Mohammad Sadoghi. 2018. GRFusion: Graphs as First-Class Citizens in Main- Memory Relational Database Systems. In Proceedings of the 2018 International Conference on Management of Data . 1789–1792. https:/...

    2018 doi

  20. [28]

    Zijin Hong, Zheng Yuan, Qinggang Zhang, Hao Chen, Junnan Dong, Feiran Huang, and Xiao Huang. 2024. Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL. arXiv preprint arXiv:2406.08426 (2024)

    2024

  21. [29]

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. 2024. GPT-4o System Card. arXiv preprint arXiv:2410.21276 (2024). https://doi.org/10. 48550/arXiv.2410.21276

    2024 doi arXiv

  22. [30]

    IBM Db2. [n.d.]. https://www.ibm.com/products/db2

  23. [31]

    Alekh Jindal, Shi Qiao, Sathwik Madhula, Kanupriya Raheja, and Sandhya Jain

  24. [32]

    Zhi Jing, Yongye Su, Yikun Han, Bo Yuan, Haiyun Xu, Chunjiang Liu, Kehai Chen, and Min Zhang. 2024. When Large Language Models Meet Vector Databases: A Survey. arXiv preprint arXiv:2402.01763 (2024). https://doi.org/10.48550/arXiv. 2402.01763

    2024 doi arXiv

  25. [33]

    Hanieh Khorashadizadeh, Fatima Zahra Amara, Morteza Ezzabady, Frédéric Ieng, Sanju Tiwari, Nandana Mihindukulasooriya, Jinghua Groppe, Soror Sahri, Farah Benamara, and Sven Groppe. 2024. Research Trends for the Interplay between Large Language Models and Knowledge Graphs. arXi...

    2024 doi arXiv

  26. [34]

    Kyoungmin Kim and Anastasia Ailamaki. 2024. Trustworthy and Efficient LLMs Meet Databases. arXiv preprint arXiv:2412.18022 (2024). https://doi.org/10.48550/ arXiv.2412.18022

    2024 doi arXiv

  27. [35]

    Jinyang Li, Binyuan Hui, Ge Qu, Jiaxi Yang, Binhua Li, Bowen Li, Bailin Wang, Bowen Qin, Ruiying Geng, Nan Huo, et al. 2023. Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to- SQLs. Advances in Neural Information Processing Sy...

    2023

  28. [36]

    Zhaodonghui Li, Haitao Yuan, Huiming Wang, Gao Cong, and Lidong Bing. 2024. LLM-R2: A Large Language Model Enhanced Rule-Based Rewrite System for Boosting Query Efficiency. Proceedings of the VLDB Endowment 18, 1 (2024), 53–65

    2024

  29. [37]

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Cheng- gang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2024. DeepSeek-V3 Technical Report. arXiv preprint arXiv:2412.19437 (2024)

    2024 arXiv

  30. [38]

    Microsoft SQL Server. [n.d.]. https://www.microsoft.com/en-us/sql-server

  31. [39]

    Mistral. [n.d.]. https://mistral.ai/

  32. [40]

    MySQL. [n.d.]. https://www.mysql.com/

  33. [41]

    Neo4j. [n.d.]. https://neo4j.com/

  34. [42]

    OpenAI: CLIP. [n.d.]. https://openai.com/index/clip/

  35. [43]

    Oracle Database. [n.d.]. https://www.oracle.com/database/

  36. [44]

    James Jie Pan, Jianguo Wang, and Guoliang Li. 2024. Survey of Vector Database Management Systems. The VLDB Journal 33, 5 (2024), 1591–1615

    2024

  37. [45]

    Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu

  38. [46]

    Simone Papicchio, Simone Rossi, Luca Cagliero, and Paolo Papotti. 2025. Think2SQL: Reinforce LLM Reasoning Capabilities for Text2SQL. arXiv preprint arXiv:2504.15077 (2025). https://doi.org/10.48550/arXiv.2504.15077

    2025 doi arXiv

  39. [47]

    Paolo Papotti. 2024. Querying Structured and Unstructured Data: LLM-first or DB-first? Keynote Talk of 29th International Conference on Database Systems for Advanced Applications (DASFAA)

    2024

  40. [48]

    IEEE Transactions on Knowledge and Data Engineering 36, 7 (2024), 3580–3599

    Unifying Large Language Models and Knowledge Graphs: A Roadmap. IEEE Transactions on Knowledge and Data Engineering 36, 7 (2024), 3580–3599. https://doi.org/10.1109/TKDE.2024.3352100

    2024

  41. [49]

    Linnea Passing, Manuel Then, Nina C Hubig, Harald Lang, Michael Schreier, Stephan Günnemann, Alfons Kemper, and Thomas Neumann. 2017. SQL-and Operator-centric Data Analytics in Relational Main-Memory Databases. In 20th International Conference on Extending Database Technology ...

    2017

  42. [50]

    Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, and Furu Wei. 2023. Kosmos-2: Grounding Multimodal Large Language Models to the World. arXiv preprint arXiv:2306.14824 (2023)

    2023 arXiv

  43. [51]

    Hyunbyung Park, Sukyung Lee, Gyoungjin Gim, Yungi Kim, Dahyun Kim, and Chanjun Park. 2024. Dataverse: Open-Source ETL (Extract, Transform, Load) Pipeline for Large Language Models. arXiv preprint arXiv:2403.19340 (2024)

    2024 arXiv

  44. [52]

    PostgreSQL. [n.d.]. https://www.postgresql.org/

  45. [53]

    Kaushik Rajan, Aseem Rastogi, Akash Lal, Sampath Rajendra, Krithika Subrama- nian, and Krut Patel. 2024. Welding Natural Language Queries to Analytics IRs with LLMs. In CIDR

    2024

  46. [54]

    Maximilian Plazotta and Meike Klettke. 2024. Data Architectures in Cloud Environments. Datenbank-Spektrum (2024), 243–247. https://doi.org/10.1007/ s13222-024-00490-5

    2024

  47. [55]

    Mohammed Saeed, Nicola De Cao, and Paolo Papotti. 2023. A DB-First Approach to Query Factual Information in LLMs. In NeurIPS 2023 Second Table Representa- tion Learning Workshop. https://openreview.net/forum?id=R8VFPAfOcN

    2023

  48. [56]

    Mohammed Saeed, Nicola De Cao, and Paolo Papotti. 2024. Querying Large Lan- guage Models with SQL. In 27th International Conference on Extending Database Technology (EDBT). 365–372. https://openproceedings.org/2024/conf/edbt/paper- 61.pdf

    2024

  49. [57]

    Sumedh Rasal. 2024. A Multi-LLM Orchestration Engine for Personalized, Context-Rich Assistance. arXiv preprint arXiv:2410.10039 (2024)

    2024 arXiv

  50. [58]

    Snowflake. [n.d.]. https://www.snowflake.com/en/

  51. [59]

    Michael Stonebraker and Andrew Pavlo. 2024. What Goes Around Comes Around... And Around... ACM Sigmod Record 53, 2 (2024), 21–37

    2024

  52. [60]

    Moritz Sichert and Thomas Neumann. 2022. User-Defined Operators: Efficiently Integrating Custom Algorithms into Modern Databases. Proceedings of the VLDB Endowment 15, 5 (2022), 1119–1131. https://doi.org/10.14778/3510397.3510408

    2022

  53. [61]

    Zhaoyan Sun, Xuanhe Zhou, and Guoliang Li. 2024. R-Bot: An LLM-based Query Rewrite System. arXiv preprint arXiv:2412.01661 (2024)

    2024 arXiv

  54. [62]

    Jie Tan, Kangfei Zhao, Rui Li, Jeffrey Xu Yu, Chengzhi Piao, Hong Cheng, Helen Meng, Deli Zhao, and Yu Rong. 2025. Can Large Language Models Be Query Optimizer for Relational Databases? arXiv preprint arXiv:2502.05562 (2025)

    2025 arXiv

  55. [63]

    Wenbo Sun, Ziyu Li, Vaishnav Srinidhi, and Rihan Hai. 2025. Database is All You Need: Serving LLMs with Relational Queries. In International Conference on Extending Database Technology (EDBT) . 1118–1121. https://openproceedings. org/2025/conf/edbt/paper-326.pdf

    2025

  56. [64]

    Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, et al. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530 (2024)

    2024 arXiv

  57. [65]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. LLaMA: Open and Efficient Foundation Language Models. arXiv preprint arXiv:2302.13971 (2023). https://do...

    2023 doi arXiv

  58. [66]

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al

  59. [67]

    Matthias Urban and Carsten Binnig. 2024. Demonstrating CAESURA: Language Models as Multi-Modal Query Planners. In Companion of the 2024 International Conference on Management of Data . 472–475. https://doi.org/10.1145/3626246. 3654732

    2024 doi

  60. [68]

    Matthias Urban and Carsten Binnig. 2024. Efficient Learned Query Execution over Text and Tables [Technical Report]. arXiv preprint arXiv:2410.22522 (2024). https://doi.org/10.48550/arXiv.2410.22522

    2024 doi arXiv

  61. [69]

    Matthias Urban and Carsten Binnig. 2024. ELEET: Efficient Learned Query Execution over Text and Tables. Proceedings of the VLDB Endowment 17, 13 (2024), 4867–4880. https://doi.org/10.14778/3704965.3704989

    2024

  62. [70]

    Matthias Urban and Carsten Binnig. 2024. CAESURA: Language Models as Multi-Modal Query Planners. In Conference on Innovative Data Systems Research (CIDR). https://vldb.org/cidrdb/papers/2024/p14-urban.pdf

    2024

  63. [71]

    Kai Waehner. 2025. The Ultimate Data Streaming Guide: Concepts, Use Cases, Industry Stories. Confluent

    2025

  64. [72]

    Jiayi Wang and Guoliang Li. 2025. AOP: Automated and Interactive LLM Pipeline Orchestration for Answering Complex Queries. In Conference on Innovative Data Systems Research (CIDR). https://vldb.org/cidrdb/papers/2025/p32-wang.pdf

    2025

  65. [73]

    Tianshu Wang, Xiaoyang Chen, Hongyu Lin, Xianpei Han, Le Sun, Hao Wang, and Zhenyu Zeng. 2025. DBCopilot: Natural Language Querying over Massive Databases via Schema Routing. In 28th International Conference on Extending Database Technology (EDBT). 707–721. https://openproceed...

    2025

  66. [74]

    Nikita Vasilenko, Alexander Demin, and Vladimir Boorlakov. 2025. Training- Free Query Optimization via LLM-Based Plan Similarity. arXiv preprint arXiv:2506.05853 (2025)

    2025 arXiv

  67. [75]

    Zhengtong Yan, Valter Uotila, and Jiaheng Lu. 2023. Join Order Selection with Deep Reinforcement Learning: Fundamentals, Techniques, and Challenges. Pro- ceedings of the VLDB Endowment 16, 12 (2023), 3882–3885

    2023

  68. [76]

    Zhiming Yao, Haoyang Li, Jing Zhang, Cuiping Li, and Hong Chen. 2025. A Query Optimization Method Utilizing Large Language Models. arXiv preprint arXiv:2503.06902 (2025)

    2025 arXiv

  69. [77]

    Fuheng Zhao, Divyakant Agrawal, and Amr El Abbadi. 2025. Hybrid Querying Over Relational Databases and Large Language Models. In Conference on Inno- vative Data Systems Research (CIDR) . https://vldb.org/cidrdb/papers/2025/p10- zhao.pdf

    2025

  70. [78]

    Wentao Wu. 2020. A Note On Operator-Level Query Execution Cost Modeling. arXiv preprint arXiv:2003.04410 (2020). https://doi.org/10.48550/arXiv.2003.04410

    2020 doi arXiv

  71. [79]

    Xinyang Zhao, Xuanhe Zhou, and Guoliang Li. 2024. Chat2Data: An Interactive Data Analysis System with RAG, Vector Databases and LLMs. Proceedings of the VLDB Endowment 17, 12 (2024), 4481–4484. https://doi.org/10.14778/3685800. 3685905

    2024 doi

  72. [80]

    Jiehan Zhou, Yang Cao, Quanbo Lu, Weishan Zhang, Xin Liu, and Weijian Ni

  73. [81]

    Jiehan Zhou, Yang Cao, Quanbo Lu, Yan Zhang, Cong Liu, Shouhua Zhang, and Junsuo Qu. 2024. Industrial Large Model: A Survey. InMATEC Web of Conferences, Vol. 401. EDP Sciences, 10009. https://doi.org/10.1051/matecconf/202440110009

    2024

  74. [82]

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al . 2023. A Survey of Large Language Models. arXiv preprint arXiv:2303.18223 (2023). https: //doi.org/10.48550/arXiv.2303.18223

    2023 doi arXiv

  75. [83]

    Xuanhe Zhou, Chengliang Chai, Guoliang Li, and Ji Sun. 2020. Database Meets Artificial Intelligence: A Survey. IEEE Transactions on Knowledge and Data Engineering 34, 3 (2020), 1096–1116

    2020

  76. [84]

    Xuanhe Zhou, Xinyang Zhao, and Guoliang Li. 2024. LLM-Enhanced Data Management. arXiv preprint arXiv:2402.02643 (2024)

    2024 arXiv

  77. [85]

    In 2024 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE)

    Industrial Large Model: Toward A Generative AI for Industry. In 2024 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE) . IEEE, 80–81

    2024

  78. [87]

    Lixi Zhou, Qi Lin, Kanchan Chowdhury, Saif Masood, Alexandre Eichenberger, Hong Min, Alexander Sim, Jie Wang, Yida Wang, Kesheng Wu, et al. 2024. Serving Deep Learning Models from Relational Databases. In27th International Conference on Extending Database Technology (EDBT) . 7...

    2024

  79. [2022]

    Advances in neural information processing systems 35 (2022), 23716–23736

    Flamingo: a Visual Language Model for Few-Shot Learning. Advances in neural information processing systems 35 (2022), 23716–23736

    2022

  80. [2023]

    arXiv preprint arXiv:2312.11805 (2023)

    Gemini: A Family of Highly Capable Multimodal Models. arXiv preprint arXiv:2312.11805 (2023)

    2023 arXiv

  81. [2024]

    In Conference on Innova- tive Data Systems Research (CIDR)

    Turning Databases Into Generative AI Machines. In Conference on Innova- tive Data Systems Research (CIDR)

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.