Pith. sign in

REVIEW 5 major objections 5 minor 60 references

TraceCAD: Trace-Guided Repair for Agentic CAD Generation

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

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

Pith's one-line read Persistent, localized, reusable repair—not another regeneration—drives reliable CAD recovery.

desk verdict Solid systems contribution to CAD repair whose central causal claim needs repeated-seed and order-permutation evidence before it lands. read the letter →

arxiv 2608.03062 v1 pith:OM5WMCJX submitted 2026-08-04 cs.AI cs.GRcs.SE

Fengxiao Fan , Jingzhe Ni , Fan Sang , Xiaolong Yin , Yu Liu , Ruofeng Tong , Min Tang , Peng Du This is my paper · ORCID
classification cs.AIcs.GRcs.SE
keywords CADgenerationprogramrepairLLMagentspersistentstatelocalizedsearchskillmemoryparametricagenticrecovery
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

TraceCAD wraps an executable CAD generation agent with a persistent recovery layer: it records which requested features are satisfied, which modeling step failed, and which repairs were tried, then uses that record to diagnose one or two faulty steps, search bounded edits in their dependency neighborhood, and validate candidates by executing them and checking that unrelated geometry is preserved. The paper claims that this persistent, localized, reusable recovery is what makes repair reliable: removing persistent state nearly halves the recovery score (from 0.9167 to 0.4872), and removing localized search more than doubles geometric regression (from 0.3456 to 0.7296). The point of caring is that CAD models are parametric programs, not just shapes: a repair that silently rewrites valid construction history is almost as bad as the original failure, and TraceCAD's design is explicitly aimed at avoiding that.

What carries the argument

The load-bearing mechanism is the persistent recovery record, which has two linked anchors: feature state F, tracking the lifecycle of each requested feature such as a body, hole pattern, edge treatment, or export, and StepTrace T, an ordered trace of instrumented code scopes each annotated with intent, execution status, and exception evidence. Repair runs on top of this record: a dependency graph built from variables defined and used by each step, a diagnosis selecting a likely faulty step, the bounded edit region R_h(s) = {s} ∪ Upstream(s, h) with hop budget h starting at one and expanding at most once to two, and a candidate selection that favors fewer changed step scopes and requires execution plus preservation evidence. Successful or failed trajectories induce reusable repair skills k = (σ, d, π, τ, ε, u)—retrieval key, transferable diagnosis, policy, patch template, evidence, and reuse statistics—so later failures are guided by prior outcomes.

What would settle it

Run TraceCAD on a suite of CAD models whose reference mismatch is deliberately spread across many steps—for instance, a boolean union whose evaluation order corrupts the entire solid, or a program whose step scopes have been scrambled so the recorded intents no longer match the code regions. If bounded two-hop search recovers almost none of these while full-program regeneration does, the localizability assumption fails and the recovery-score gains vanish.

Watch

Extended reading notes

Core claim

On its own terms, TraceCAD's central claim is that a CAD agent's correction loop should be backed by an explicit, persistent record—linking requested features, modeling steps, failure evidence, and candidate outcomes—so that recovery becomes a localized, evidence-backed edit rather than another unconstrained generation attempt. Concretely, the paper defines feature state F and a StepTrace T over instrumented code scopes, diagnoses a likely faulty step, builds a repair region R_h(s) = {s} ∪ Upstream(s, h) starting at one hop and expanding at most once to two, and promotes a patch only when execution, artifact export, semantic and visual agreement, preservation of unrelated geometry, and edit locality all pass. The ablation results are the direct support: the full cold-start system reaches a Recovery Score of 0.9167 with Geometric Regression 0.3456 and a Repair Scope Ratio of 0.3899, whereas removing persistent state drops recovery to 0.4872 and removing localized search raises regression to 0.7296 and average retries to 3.0. The paper concludes that persistent, localized, and reusable recovery improves both final CAD geometry and repair reliability, and that this holds across LLM backends while reducing token cost and latency when the skill store is warm-started on disjoint training models.

Load-bearing premise

The load-bearing premise is that residual CAD failures are local: the faulty operation can be traced to one or two modeling steps within a small dependency neighborhood, and patching that neighborhood fixes the model without disturbing already-correct geometry.

Editorial extensions

If this is right

  • If TraceCAD's central claim is right, then recovery quality in agentic CAD should be measured not by final shape alone but by whether the intended defect was fixed without rewriting valid construction history; Recovery Score and Geometric Regression capture that distinction.
  • Keeping feature and step evidence across attempts is what enables reliable recovery: the state ablation shows Recovery Score falling from 0.9167 to 0.4872 when that state is removed.
  • Bounding repairs to a diagnosed step's dependency neighborhood prevents broad, damaging rewrites: without localized search, Geometric Regression rises from 0.3456 to 0.7296 and average code-agent invocations double to 3.0.
  • Reusing successful and failed repair outcomes in a skill store cuts retries, token cost, and latency: warm-starting on 1K disjoint training models lowers average retries from 1.5 to 1.2 and tokens from 103.6K to 70.3K.
  • The recovery layer is generator-agnostic: it wraps initial generation without changing decoding, so gains from persistent state, localized search, and skill memory should transfer to other executable CAD generators.

Reading between the lines

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

  • The paper's localizability premise could be stress-tested by engineering failures that are truly global—for example, a boolean operation whose effect corrupts the entire solid—and measuring whether bounded two-hop search still recovers them; the paper itself notes that poorly decomposed code makes step attribution ambiguous.
  • The same persistent-state recipe—feature obligations, step traces, bounded dependency search, and skill memory—could apply to any agent that produces structured executable artifacts such as scripts, assembly plans, or shader programs, whenever a repair must preserve already-correct parts.
  • Recovery Score, defined as success weighted by the inverse number of repair invocations, is a natural general metric for agentic repair; adopting it across code-generation benchmarks would make repair reliability comparable across systems.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. TraceCAD proposes a persistent recovery layer for LLM-based CAD agents. The system maintains feature and step state across the correction loop, diagnoses a likely faulty modeling step from execution/visual evidence, generates bounded local patches in a dependency region, and accumulates reusable repair skills with reuse statistics. The paper evaluates TraceCAD on DeepCAD-derived 200-model and 1K-model subsets, reporting geometric fidelity (IoU, Chamfer distance, Hausdorff distance), repair behavior (Recovery Score, Regression, Scope, Reuse), and efficiency (retry count, tokens, latency). The central claim is that persistent state, localized search, and skill memory improve final CAD quality and repair reliability, supported mainly by the ablation contrasts in Table 2.

Significance. The paper addresses a genuine problem in agentic CAD generation: correction loops lose evidence about satisfied requirements, faulty operations, and prior repairs, leading to regressions and inefficient retries. The proposed recovery layer — persistent feature/step state, bounded localized repair search, and reuse-aware skill memory — is a well-motivated and potentially reusable design. The evaluation goes beyond final geometry by introducing repair-specific metrics (Recovery Score, Regression, Scope, Reuse), and the supplement provides detailed metric definitions, backend sensitivity, cost comparisons, and paired Wilcoxon tests for baseline comparisons. If the ablation results are robust, the system would be a useful contribution to CAD agents and to agentic program repair more broadly. However, the central causal attribution currently rests on single-run ablations with no uncertainty quantification, and the paper itself acknowledges order dependence from online skill accumulation. These issues must be addressed before the main claim is fully supported.

major comments (5)
  1. [§Experiments (Ablations), Table 2, and Supp. B.2] The central claim rests on single-run ablation contrasts with no uncertainty quantification. Supp. B.2 states that each configuration is executed once per selected case and that hosted LLM endpoints do not provide a consistently enforceable sampling seed, so generation remains stochastic. Since the cold-start skill store accumulates from earlier evaluation cases and can be reused later (as the Limitations section explicitly notes), the reported differences (e.g., Recovery 0.9167 vs 0.4872; Regression 0.3456 vs 0.7296) are one trajectory of an order-dependent stochastic process. The paper should report repeated runs (e.g., multiple seeds and order permutations) with confidence intervals, or at least per-case paired tests for the ablations analogous to the Wilcoxon tests in Supp. B.5, before attributing the differences to the ablated components.
  2. [§A.2 (Eq. 7) and Table 2] The Recovery Score is averaged only over cases whose initial generation fails, and the Regression metric only over cases with distinct pre- and post-repair artifacts, but the paper does not report the cardinalities |F| and |V_Reg|. If the initial failure rate is low, the stated Recovery differences may be based on a very small number of cases, making them fragile. The paper should report these denominators and, ideally, the per-case distribution of recovery outcomes for every variant in Table 2.
  3. [§Comparison with Existing Methods, Table 4, and Supp. B.5] The text says TraceCAD 'achieves the strongest overall geometric fidelity' and frames the comparison with CADDesigner favorably, but the paired statistics in Table 7 show the IoU difference from CADDesigner is not significant (Holm-adjusted p=0.7884), and the HD difference is also not significant (p=0.7884). Only the Text2CAD differences are significant after correction. The paper should either weaken the wording to 'comparable' or provide additional evidence before claiming superiority over the strongest baseline.
  4. [§Method (Localized Repair Search, Eq. 2) and §Setup] The locality hyperparameters — hop budget h (starting at 1, expanding to 2), at most three candidates over at most two target steps, visual-score threshold of five points, and skill retrieval top-5 — are fixed with no sensitivity analysis. Because the central mechanism is bounded local search, the reader cannot tell whether the benefit is robust to the repair-region size or is tuned to the reported setting. The paper should include an ablation over h and beta (e.g., h=0, 1, 2, 3, or unrestricted), and report how often the bounded search exhausts its region or fails to find a local candidate. This would also partly address the acknowledged localizability limitation.
  5. [§Ablations, Table 2 (w/o visual feedback)] The visual-feedback ablation removes rendered semantic evidence from diagnosis and repair validation, but it also changes the termination and promotion rule: without visual evidence, the run stops as soon as execution and export succeed, which lowers AVG Re (1.4), tokens, and latency. The comparison therefore conflates the absence of visual evidence with a different retry budget. To isolate the contribution of visual feedback, the paper should hold the retry budget fixed (or report results as a function of budget) so that the lower cost reflects the evidence source rather than an earlier stopping policy.
minor comments (5)
  1. [Table 2 caption] A dash is said to indicate absent evidence, but for the 'w/o localized repair search' row the absence of Scope and Reuse should be explained more explicitly, since full-program repairs still produce promoted candidates in this variant.
  2. [§A.2 (Eq. 7) and §A.3 (Eq. 13)] a_i in Eq. (7) is defined as the number of subsequent code-agent invocations, whereas AVG Re in Eq. (13) counts the initial generation as one invocation; the paper should state this distinction clearly in one place to avoid misinterpretation of the two metrics.
  3. [§Setup and Table 5] The phrase 'retry budget' is used as an umbrella term, but the individual limits in Table 5 (30 main-agent tool steps, 20 internal steps, 180-second timeout, five consecutive failures) are numerous; a brief sentence connecting these to Eq. (1)'s beta would help the reader understand the search budget.
  4. [Introduction] The claim that 'many failures are local even when their visual effects are global' is plausible but is not referenced to prior CAD or program-repair literature; a citation or a small empirical count from the failures observed in this study would strengthen it.
  5. [Figure 2 and Algorithm 1] Figure 2 draws skill retrieval before diagnosis, which is consistent with Algorithm 1 (line 7 retrieves skills, line 8 diagnoses), but the prose in §Method describes diagnosis before skill retrieval; align the presentation order in the text.

Circularity Check

0 steps flagged · score 2.0 of 10

No meaningful circularity; TraceCAD's recovery gains are measured against external benchmarks and independent metric definitions, with only non-load-bearing same-author citations.

full rationale

TraceCAD's central claim, that persistent, localized, and reusable recovery improves final CAD quality and repair reliability, is supported by ablations on DeepCAD-derived benchmarks. The geometric metrics (IoU, Chamfer distance, Hausdorff distance) and repair metrics (Recovery Score, Regression, Scope, Reuse) are defined in Supplement A from reference solids, voxel comparisons, invocation counts, step-scope diffs, and retrieval outcomes, not from TraceCAD's internal state or fitted parameters. There is no fitted quantity that is later reported as a prediction, and no equation in the derivation reduces to its own inputs by construction. The same-author citation to CADDesigner (Fan et al. 2026) supplies the SimpleCADAPI grounding and benchmark protocol, but it does not supply the recovery results, so it is not load-bearing for the claimed contribution. Warm-up skill initialization uses disjoint training models, so the efficiency comparison is not circular, though it does share the DeepCAD distribution. The acknowledged single-run stochastic evaluation and cold-start order sensitivity (Supplementary B.2; Limitations) are statistical-validity concerns rather than circularity: they do not make any measured outcome equivalent to an input by definition. The localizability assumption is explicit rather than smuggled in. Overall, the evaluation is externally grounded and the central claims do not reduce to self-citation or definitional identity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

No new physical or theoretical entities are proposed. Persistent recovery state, StepTrace, and repair skills are implemented software data structures and abstractions, not unobserved entities with independent falsifiable predictions. The central claim rests on hand-chosen search budgets, meaningful step instrumentation, the reliability of learned visual feedback, and single-run stochastic evaluation.

free parameters (5)
  • Hop budget h for repair region = starts at 1 upstream hop, may expand once to 2 hops
    Controls how localized the repair search is in Eq. (2) and the implementation settings. The paper's central claim about localized search depends on this hand-chosen budget, but no sensitivity analysis is reported.
  • Candidate and target-step budget beta = at most 3 candidates over at most 2 target steps
    Bounds the search in Algorithm 1 and Table 5. It directly affects Recovery Score, token cost, and regression, but the paper does not justify or vary it.
  • Visual promotion threshold = passing shape-delta judgment with no unintended changes, or visual-score improvement of at least 5 points
    Hand-set promotion rule in Experiments. Because success and repair selection depend on this threshold, the threshold choice matters for all headline results.
  • Skill retrieval top-k = at most 5 skill matches
    Implementation setting; affects Reuse Precision and token cost, with no sensitivity analysis reported.
  • Retry budget limits = 3 deterministic fast-repair attempts, 2 complex-repair invocations, termination after 5 consecutive unsuccessful…
    Hand-set budgets in Supplement B.2. Recovery Score and AVG Re are defined through these limits, so results are conditional on them.
assumptions (5)
  • domain assumption Residual CAD failures can be localized to a small dependency neighborhood around a faulty modeling step.
    Introduction states 'many failures are local even when their visual effects are global'; Eq. (2) restricts repairs to one or two upstream hops. If global interactions or missing features dominate, bounded search cannot find the fault.
  • domain assumption Generated programs are decomposed into meaningful, correctly instrumented step scopes.
    StepTrace in Definition 1 and the cad_step instrumentation assume each feature maps to identifiable code regions. The Limitations section concedes 'poorly decomposed code makes step attribution ambiguous.'
  • domain assumption The vision model's semantic judgments and shape-delta checks are reliable enough to gate success and promotion.
    The System section says visual agreement is 'heuristic rather than a formal geometric guarantee'; promotion requires visual improvement or no unintended changes, so every success claim passes through this learned gate.
  • domain assumption One evaluation run per configuration, without enforceable LLM seeds, is representative of performance.
    Supplement B.2 states hosted endpoints do not provide enforceable sampling seeds and each configuration is executed once per case. The headline tables have no error bars.
  • domain assumption DeepCAD-derived subsets with GPT-5.5 shape descriptions are a valid benchmark for text-CAD repair.
    Supplement B.1 constructs 200 and 1K subsets from DeepCAD, generates descriptions from fixed-view renderings, and substitutes unexportable cases with the next seeded candidate; the conclusions assume this setup reflects real text-to-CAD use.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TraceCAD: Trace-Guided Repair for Agentic CAD Generation." pith.science (2026). https://pith.science/paper/OM5WMCJX

@misc{pith2026260803062,
  author       = {Pith},
  title        = {Pith review of: TraceCAD: Trace-Guided Repair for Agentic CAD Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OM5WMCJX}},
  note         = {Machine review of arXiv:2608.03062}
}
read the original abstract

LLM-based CAD agents produce executable parametric programs, but their correction loops may lose evidence about satisfied requirements, faulty operations, and prior repairs. We introduce TraceCAD, a recovery layer that links requested features, modeling steps, failure evidence, and candidate outcomes as persistent state. TraceCAD diagnoses likely faulty operations, searches bounded edits in their dependency regions, validates candidates through execution and preservation checks, and retains successful and failed repair outcomes in reusable skill memory. On DeepCAD-derived benchmarks with 200-model ablations and a 1K-model comparison, TraceCAD achieves competitive geometric quality in terms of IoU, Chamfer distance, and Hausdorff distance. Removing persistent state nearly halves recovery score; removing localized search more than doubles geometric regression and doubles code-agent invocations. Initializing the skill store on disjoint training models further reduces retries, token cost, and latency. These results demonstrate that persistent, localized, and reusable recovery improves final CAD quality and repair reliability.

Figures

Figures reproduced from arXiv: 2608.03062 by the authors.

Figure 1
Figure 1. Representative CAD outputs produced by TraceCAD, spanning diverse mechanical parts and multi-component [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of TraceCAD. Given a text or multimodal CAD requirement, the agent refines the request, generates [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Repair-skill lifecycle. A failure trajectory is di [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative results under identical prompts for a [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Average token usage and end-to-end latency on the 200-model subset. Each metric is normalized independently by its [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Image-conditioned generation and localized repair of the geared electric motor: (a) conditioning sketch, (b) failure [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: The validated geared-motor model reconstructed in FreeCAD. Although collapsed in the screenshot, the tree at left [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 56 canonical work pages

  1. [1]

    H.; Langwasser, M.; Xu, W.; and Gombolay, M

    Alrashedy, K.; Tambwekar, P.; Zaidi, Z. H.; Langwasser, M.; Xu, W.; and Gombolay, M. 2025. Generating CAD Code with Vision-Language Models for 3D Designs. In International Conference on Learning Representations

    work page 2025

  2. [3]

    Chen, C.; Wei, J.; Chen, T.; Zhang, C.; Yang, X.; Zhang, S.; Yang, B.; Foo, C.-S.; Lin, G.; Huang, Q.; and Liu, F. 2025. CADCrafter : Generating Computer-Aided Design Models from Unconstrained Images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    work page 2025

  3. [4]

    Fan, F.; Ni, J.; Yin, X.; Wang, S.; Lu, X.; Zou, Q.; Tong, R.; Tang, M.; and Du, P. 2026. CADDesigner : Conceptual CAD Model Generation with a General-Purpose Agent. Computer-Aided Design, 198: 104087

    work page 2026

  4. [5]

    Govindarajan, P.; Baldelli, D.; Pathak, J.; Fournier, Q.; and Chandar, S. 2026. CADmium : Fine-Tuning Code Language Models for Text-Driven Sequential CAD Design. Transactions on Machine Learning Research

    work page 2026

  5. [6]

    Guan, Y.; Wang, X.; Xing, X.; Zhang, J.; Xu, D.; and Yu, Q. 2025. CAD-Coder : Text-to- CAD Generation with Chain-of-Thought and Geometric Reward. In Advances in Neural Information Processing Systems, volume 38

    work page 2025

  6. [7]

    Guo, H.; Liu, S.; Pan, H.; Liu, Y.; Tong, X.; and Guo, B. 2022. ComplexGen : CAD Reconstruction by B-rep Chain Complex Generation. ACM Transactions on Graphics, 41(4): 1--18

    work page 2022

  7. [9]

    S.; Sinha, S.; Sheikh, T

    Khan, M. S.; Sinha, S.; Sheikh, T. U.; Stricker, D.; Ali, S. A.; and Afzal, M. Z. 2024. Text2CAD : Generating Sequential CAD Designs from Beginner-to-Expert Level Text Prompts. In Advances in Neural Information Processing Systems, volume 37, 7552--7579

    work page 2024

  8. [10]

    Le Goues, C.; Nguyen, T.; Forrest, S.; and Weimer, W. 2012. GenProg : A Generic Method for Automatic Software Repair. IEEE Transactions on Software Engineering, 38(1): 54--72

    work page 2012

Show all 60 references
  1. [11]

    Li, J.; Ma, W.; Li, X.; Lou, Y.; Zhou, G.; and Zhou, X. 2025 a . CAD-Llama : Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18563--18573

    2025

  2. [12]

    Li, X.; Li, J.; Song, Y.; Lou, Y.; and Zhou, X. 2026. Seek-CAD : A Self-Refined Generative Modeling for 3D Parametric CAD Using Local Inference via DeepSeek . In International Conference on Learning Representations

    2026

  3. [13]

    Li, Y.; Lin, C.; Liu, Y.; Long, X.; Zhang, C.; Wang, N.; Li, X.; Wang, W.; and Guo, X. 2025 b . CADDreamer : CAD Object Generation from Single-View Images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    2025

  4. [15]

    Liu, K.; Koyuncu, A.; Kim, D.; and Bissyande, T. F. 2019. TBar : Revisiting Template-Based Automated Program Repair. In Proceedings of the 28th ACM SIGSOFT International Symposium on Software Testing and Analysis , 31--42

    2019

  5. [16]

    Ma, W.; Chen, S.; Lou, Y.; Li, X.; and Zhou, X. 2024. Draw Step by Step: Reconstructing CAD Construction Sequences from Point Clouds via Multimodal Diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 27154--27163

    2024

  6. [17]

    Madaan, A.; Tandon, N.; Gupta, P.; Hallinan, S.; Gao, L.; Wiegreffe, S.; Alon, U.; Dziri, N.; Prabhumoye, S.; Yang, Y.; et al. 2023. Self-Refine: Iterative Refinement with Self-Feedback. In Advances in Neural Information Processing Systems, volume 36, 46534--46594

    2023

  7. [18]

    S.; Cavada, S.; Rukhovich, D.; Foteinopoulou, N.; Cherenkova, K.; Kacem, A.; and Aouada, D

    Mallis, D.; Karadeniz, A. S.; Cavada, S.; Rukhovich, D.; Foteinopoulou, N.; Cherenkova, K.; Kacem, A.; and Aouada, D. 2025. CAD-Assistant : Tool-Augmented VLLMs as Generic CAD Task Solvers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 7284--7294

    2025

  8. [19]

    Rukhovich, D.; Dupont, E.; Mallis, D.; Cherenkova, K.; Kacem, A.; and Aouada, D. 2025. CAD-Recode : Reverse Engineering CAD Code from Point Clouds. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 9801--9811

    2025

  9. [20]

    Wang, R.; Yuan, Y.; Sun, S.; and Bian, J. 2025. Text-to- CAD Generation Through Infusing Visual Feedback in Large Language Models. In Proceedings of the 42nd International Conference on Machine Learning, volume 267 of Proceedings of Machine Learning Research, 65326--65345

    2025

  10. [21]

    Willis, K. D. D.; Pu, Y.; Luo, J.; Chu, H.; Du, T.; Lambourne, J. G.; Solar-Lezama, A.; and Matusik, W. 2021. Fusion 360 Gallery: A Dataset and Environment for Programmatic CAD Construction from Human Design Sequences. ACM Transactions on Graphics, 40(4)

    2021

  11. [22]

    Wu, R.; Xiao, C.; and Zheng, C. 2021. DeepCAD : A Deep Generative Network for Computer-Aided Design Models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 6772--6782

    2021

  12. [23]

    S.; Wei, Y.; and Zhang, L

    Xia, C. S.; Wei, Y.; and Zhang, L. 2023. Automated Program Repair in the Era of Large Pre-Trained Language Models. In Proceedings of the 45th IEEE / ACM International Conference on Software Engineering , 1482--1494

    2023

  13. [24]

    K.; Lambourne, J

    Xu, X.; Jayaraman, P. K.; Lambourne, J. G.; Willis, K. D. D.; and Furukawa, Y. 2023. Hierarchical Neural Coding for Controllable CAD Model Generation. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Researc...

    2023

  14. [25]

    Xu, X.; Lambourne, J.; Jayaraman, P.; Wang, Z.; Willis, K.; and Furukawa, Y. 2024. BrepGen : A B-rep Generative Diffusion Model with Structured Latent Geometry. ACM Transactions on Graphics, 43(4): 1--14

    2024

  15. [26]

    Xu, X.; Willis, K. D. D.; Lambourne, J. G.; Cheng, C.-Y.; Jayaraman, P. K.; and Furukawa, Y. 2022. SkexGen : Autoregressive Generation of CAD Construction Sequences with Disentangled Codebooks. In International Conference on Machine Learning, 24698--24724

    2022

  16. [28]

    R.; and Cao, Y

    Yao, S.; Zhao, J.; Yu, D.; Du, N.; Shafran, I.; Narasimhan, K. R.; and Cao, Y. 2023. ReAct : Synergizing Reasoning and Acting in Language Models. In International Conference on Learning Representations

    2023

  17. [29]

    Zhang, A.; Jia, W.; Zou, Q.; Feng, Y.; Wei, X.; and Zhang, Y. 2025 a . Diffusion-CAD : Controllable Diffusion Model for Generating Computer-Aided Design Models. IEEE Transactions on Visualization and Computer Graphics

    2025

  18. [30]

    Zhang, Z.; Sun, S.; Wang, W.; Cai, D.; and Bian, J. 2025 b . FlexCAD : Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models. In International Conference on Learning Representations

    2025

  19. [31]

    D.; and Company, P

    Zhou, J.; Camba, J. D.; and Company, P. 2026. CADialogue : A Multimodal LLM -Powered Conversational Assistant for Intuitive Parametric CAD Modeling. Computer-Aided Design, 191: 104006

    2026

  20. [32]

    Zhou, J.; Gupta, D.; and Camba, J. D. 2026. Prompt2CAD : A Lightweight LLM Framework for Conversational CAD Generation and Iterative Refinement. Procedia Computer Science, 277: 1663--1672

    2026

  21. [33]

    2024 , url=

    Khan, Mohammad Sadil and Sinha, Sankalp and Sheikh, Talha Uddin and Stricker, Didier and Ali, Sk Aziz and Afzal, Muhammad Zeshan , booktitle=. 2024 , url=

    2024

  22. [34]

    2025 , url=

    Zhang, Zhanwei and Sun, Shizhao and Wang, Wenxiao and Cai, Deng and Bian, Jiang , booktitle=. 2025 , url=

    2025

  23. [35]

    Text-to-

    Wang, Ruiyu and Yuan, Yu and Sun, Shizhao and Bian, Jiang , booktitle=. Text-to-. 2025 , url=

    2025

  24. [36]

    2025 , url=

    Li, Jiahao and Ma, Weijian and Li, Xueyang and Lou, Yunzhong and Zhou, Guichun and Zhou, Xiangdong , booktitle=. 2025 , url=

    2025

  25. [37]

    2026 , url=

    Govindarajan, Prashant and Baldelli, Davide and Pathak, Jay and Fournier, Quentin and Chandar, Sarath , journal=. 2026 , url=

    2026

  26. [38]

    2025 , url=

    Guan, Yandong and Wang, Xilin and Xing, Ximing and Zhang, Jing and Xu, Dong and Yu, Qian , booktitle=. 2025 , url=

    2025

  27. [39]

    Generating

    Alrashedy, Kamel and Tambwekar, Pradyumna and Zaidi, Zulfiqar Haider and Langwasser, Megan and Xu, Wei and Gombolay, Matthew , booktitle=. Generating. 2025 , url=

    2025

  28. [40]

    , journal=

    Zhou, Jiwei and Gupta, Deepanshu and Camba, Jorge D. , journal=. 2026 , doi=

    2026

  29. [41]

    and Company, Pedro , journal=

    Zhou, Jiwei and Camba, Jorge D. and Company, Pedro , journal=. 2026 , doi=

    2026

  30. [42]

    2603.26512 , archivePrefix=

    Barkley, Jesse and Loghmani, Rumi and Barati Farimani, Amir , year=. 2603.26512 , archivePrefix=

  31. [43]

    2012 , doi=

    Le Goues, Claire and Nguyen, ThanhVu and Forrest, Stephanie and Weimer, Westley , journal=. 2012 , doi=

    2012

  32. [44]

    , booktitle=

    Liu, Kui and Koyuncu, Anil and Kim, Dongsun and Bissyande, Tegawende F. , booktitle=. 2019 , doi=

    2019

  33. [45]

    Proceedings of the 45th

    Automated Program Repair in the Era of Large Pre-Trained Language Models , author=. Proceedings of the 45th. 2023 , doi=

    2023

  34. [46]

    Willis, Karl D. D. and Pu, Yewen and Luo, Jieliang and Chu, Hang and Du, Tao and Lambourne, Joseph G. and Solar-Lezama, Armando and Matusik, Wojciech , journal=. Fusion 360 Gallery: A Dataset and Environment for Programmatic. 2021 , doi=

    2021

  35. [47]

    Wu, Rundi and Xiao, Chang and Zheng, Changxi , booktitle=

  36. [48]

    Xu, Xiang and Willis, Karl D. D. and Lambourne, Joseph G. and Cheng, Chin-Yi and Jayaraman, Pradeep Kumar and Furukawa, Yasutaka , booktitle=

  37. [49]

    and Willis, Karl D

    Xu, Xiang and Jayaraman, Pradeep Kumar and Lambourne, Joseph G. and Willis, Karl D. D. and Furukawa, Yasutaka , booktitle=. Hierarchical Neural Coding for Controllable. 2023 , url=

    2023

  38. [50]

    2022 , doi=

    Guo, Haoxiang and Liu, Shilin and Pan, Hao and Liu, Yang and Tong, Xin and Guo, Baining , journal=. 2022 , doi=

    2022

  39. [51]

    Draw Step by Step: Reconstructing

    Ma, Weijian and Chen, Shuaiqi and Lou, Yunzhong and Li, Xueyang and Zhou, Xiangdong , booktitle=. Draw Step by Step: Reconstructing

  40. [52]

    Xu, Xiang and Lambourne, Joseph and Jayaraman, Pradeep and Wang, Zhengqing and Willis, Karl and Furukawa, Yasutaka , journal=

  41. [53]

    Zhang, Aijia and Jia, Weiqiang and Zou, Qiang and Feng, Yixiong and Wei, Xiaoxiang and Zhang, Ye , journal=

  42. [54]

    Li, Yuan and Lin, Cheng and Liu, Yuan and Long, Xiaoxiao and Zhang, Chenxu and Wang, Ningna and Li, Xin and Wang, Wenping and Guo, Xiaohu , booktitle=

  43. [55]

    Chen, Cheng and Wei, Jiacheng and Chen, Tianrun and Zhang, Chi and Yang, Xiaofeng and Zhang, Shangzhan and Yang, Bingchen and Foo, Chuan-Sheng and Lin, Guosheng and Huang, Qixing and Liu, Fayao , booktitle=

  44. [56]

    2025 , url=

    Rukhovich, Danila and Dupont, Elona and Mallis, Dimitrios and Cherenkova, Kseniya and Kacem, Anis and Aouada, Djamila , booktitle=. 2025 , url=

    2025

  45. [57]

    2025 , url=

    Mallis, Dimitrios and Karadeniz, Ahmet Serda and Cavada, Sebastian and Rukhovich, Danila and Foteinopoulou, Niki and Cherenkova, Kseniya and Kacem, Anis and Aouada, Djamila , booktitle=. 2025 , url=

    2025

  46. [58]

    2026 , url=

    Li, Xueyang and Li, Jiahao and Song, Yu and Lou, Yunzhong and Zhou, Xiangdong , booktitle=. 2026 , url=

    2026

  47. [59]

    2026 , url=

    Fan, Fengxiao and Ni, Jingzhe and Yin, Xiaolong and Wang, Sirui and Lu, Xingyu and Zou, Qiang and Tong, Ruofeng and Tang, Min and Du, Peng , journal=. 2026 , url=

    2026

  48. [60]

    2606.13368 , archivePrefix=

    Hu, Tao and Ai, Jiaxin and Wen, Licheng and Li, Xueheng and Zou, Shu and Li, Siqi and Deng, Nianchen and Cai, Xinyu and Zhou, Hongbin and Cai, Pinlong and Fu, Daocheng and Yang, Yu and Zhang, Hairong and Shi, Botian and Yang, Xuemeng , year=. 2606.13368 , archivePrefix=

    arXiv

  49. [61]

    2606.31252 , archivePrefix=

    Liu, Fumin and Zhou, Haoyu and Hao, Fei and Yang, Lin , year=. 2606.31252 , archivePrefix=

    arXiv

  50. [62]

    2607.05750 , archivePrefix=

    Xu, Yunhan and Wu, Qifeng and Li, Xunjin and Bin, Yuanwei and Yao, Qingsong and Gu, Jianghang and Wang, Guan and Lv, Weihao and Yang, Huiyu and Luo, Wenfa and Xiang, Jiao and Chen, Yuntian and Chen, Shiyi , year=. 2607.05750 , archivePrefix=

    arXiv

  51. [63]

    and Cao, Yuan , booktitle=

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik R. and Cao, Yuan , booktitle=. 2023 , url=

    2023

  52. [64]

    Advances in Neural Information Processing Systems , volume=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. Advances in Neural Information Processing Systems , volume=

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.