EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

LLM-Compatible Visual Brief Cards for AI Infrastructure Capacity Dashboards: A UI/UX Framework for Turning Forecast Risk into Graphic Design Decisions

Ge Liu, Shilu He, Hailey Wong

International Journal of Graphic Design · 30.05.2025 · 3 (1) · 196-213

DOI: 10.51903/ijgd.v3i1.3723 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Graphic Design, online erschienen am 30.05.2025. Verfasst von Ge Liu und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

AI infrastructure dashboards frequently condense GPU inventory, demand forecasts, uncertainty, and admission-control policies into dense operational reports, making it difficult for planners to identify critical risks and appropriate design actions. This study introduces VB-Card, an LLM-compatible UI/UX framework that converts capacity-forecast evidence into structured visual brief cards containing risk indicators, capacity summaries, uncertainty cues, inventory evidence, policy explanations, and actionable recommendations. To evaluate the framework independently of vendor-specific model variation, a deterministic visual-brief generator was employed. The evaluation integrates the OpenCOLE graphic-design field contract with the Alibaba PAI GPU cluster trace. GPU demand, inventory, rolling P50/P90 forecasts, uncertainty measures, and stock-out probabilities were derived from 1,037,084 GPU-requesting task records and 1,897 machine specification records, then paired with 23,419 OpenCOLE instances. Five approaches were compared: a baseline text report, a generic visual card, an OpenCOLE keyword brief, a risk-badge card, and VB-Card. On the held-out test set, VB-Card achieved the highest overall score (0.820; 95% bootstrap CI: 0.820–0.821), outperforming the baseline text report (0.633), OpenCOLE keyword brief (0.522), and risk-badge card (0.538). Ablation results show that uncertainty, policy context, inventory evidence, and visual hierarchy each contribute to performance. The findings demonstrate that capacity risk can be translated into consistent, data-grounded visual-brief decisions, although real-world UI effectiveness requires user-based validation.

Zitieren

APA

Liu, G., He, S., & Wong, H. (2025). LLM-Compatible Visual Brief Cards for AI Infrastructure Capacity Dashboards: A UI/UX Framework for Turning Forecast Risk into Graphic Design Decisions. International Journal of Graphic Design, 3(1), 196-213. https://doi.org/10.51903/ijgd.v3i1.3723

Vancouver

Liu G, He S, Wong H. LLM-Compatible Visual Brief Cards for AI Infrastructure Capacity Dashboards: A UI/UX Framework for Turning Forecast Risk into Graphic Design Decisions. International Journal of Graphic Design. 2025;3(1):196-213. doi:10.51903/ijgd.v3i1.3723

BibTeX
@article{Liu2025,
  author = {Liu, Ge and He, Shilu and Wong, Hailey},
  title = {LLM-Compatible Visual Brief Cards for AI Infrastructure Capacity Dashboards: A UI/UX Framework for Turning Forecast Risk into Graphic Design Decisions},
  journal = {International Journal of Graphic Design},
  year = {2025},
  volume = {3},
  number = {1},
  pages = {196--213},
  doi = {10.51903/ijgd.v3i1.3723}
}
Kostenlos starten →