There is a growing interest in leveraging multiple large language models (LLMs) for automated code optimization. However, industrial platforms deploying multiple LLMs face a critical challenge: prompts optimized for one LLM often fail with others, requiring expensive model-specific prompt engineering. This cross-model prompt engineering bottleneck severely limits the practical deployment of multi-LLM systems in production environments. We introduce Meta-Prompted Code Optimization (Mpco), a framework that automatically generates high-quality, task-specific prompts across diverse LLMs while maintaining industrial efficiency requirements. Mpco leverages meta-prompting to dynamically synthesize context-aware optimization prompts by integrating project metadata, task requirements, and LLM-specific contexts. It is an essential part of the ARTEMIS code optimization platform for automated validation and scaling.Our comprehensive evaluation on five real-world codebases with 366 hours of runtime benchmarking demonstrates Mpco’s effectiveness: it achieves overall performance improvements up to 19.06% with the best statistical rank across all systems compared to baseline methods. Analysis shows that 96% of the top-performing optimizations stem from meaningful edits. Through systematic ablation studies and meta-prompter sensitivity analysis, we identify that comprehensive context integration is essential for effective meta-prompting and that major LLMs can serve effectively as meta-prompters, providing actionable insights for industrial practitioners.

Glyphs representing complex behavior provide a useful and common means of visualizing multivariate data. However, due to their complex shape, overlapping, and occlusion of glyphs is a common and prominent limitation. This limits the number of discreet data tuples that can be displayed in a given image. Using a real-world application, glyphs are used to depict agent behavior in a call center. However, many call centers feature thousands of agents. A standard approach representing thousands of agents with glyphs does not scale. To accommodate the visualization incorporating thousands of glyphs we develop clustering of overlapping glyphs into a single parent glyph. This hierarchical glyph represents the mean value of all child agent glyphs, removing overlap and reduTcing visual clutter. Multi-variate clustering techniques are explored and developed in collaboration with domain experts in the call center industry. We implement dynamic control of glyph clusters according to zoom level and customized distance metrics, to utilize image space with reduced overplotting and cluttering. We demonstrate our technique with examples and a usage scenario using real-world call-center data to visualize thousands of call center agents, revealing insight into their behavior and reporting feedback from expert call-center analysts.

Chord diagrams are a popular high-dimensional method for showing connections between nodes, however they have scalability limitations and lack advanced methods for interaction and multivariate links. In this paper we introduce a novel interaction and layout method for improving the scalability of chord diagrams, explore sketch-based methods for showing multiple links and direction, and introduce a sketch-based brushing technique for filtering. We demonstrate the interaction techniques on a real-world call-center dataset and report feedback from domain experts in the call center industry.

The Parallel Coordinates plot is a popular tool for the visualization of high-dimensional data. One of the main challenges when using parallel coordinates is occlusion and overplotting resulting from large data sets. Brushing is a popular approach to address these challenges. Since its conception, limited improvements have been made to brushing both in the form of visual design and functional interaction. We present a set of novel, smart brushing techniques that enhance the standard interactive brushing of a parallel coordinates plot. We introduce two new interaction concepts: Higher-order, sketch-based brushing, and smart, data-driven brushing. Higher-order brushes support interactive, flexible, n-dimensional pattern searches involving an arbitrary number of dimensions. Smart, data-driven brushing provides interactive, real-time guidance to the user during the brushing process based on derived meta-data. In addition, we implement a selection of novel enhancements and user options that complement the two techniques as well as enhance the exploration and analytical ability of the user. We demonstrate the utility and evaluate the results using a case study with a large, high-dimensional, real-world telecommunication data set and we report domain expert feedback from the data suppliers.