The work proposes an autonomous, fully local framework for synthesizing compression-ratio surrogates for error-controlled lossy compressors. Building on the SECRE surrogate-ratio estimation paradigm, we address the main practical bottleneck: surrogate construction typically requires compressor-specific investigation and manual engineering. Given a set of candidate compressors and an error-control specification, a large-language-model agent (i) derives compressor-compatible sampling strategies, (ii) generates an executable surrogate to estimate compressed size without full compression, and (iii) iteratively refines it via a closed-loop validator that compares predictions against short, local compression trials on held-out samples. The refinement loop reuses execution traces and prior design decisions as few-shot in-context exemplars, improving stability and reducing repeated exploration. All compilation, profiling, and testing are performed locally using the Model Context Protocol to invoke tool calls, producing deterministic and reproducible artifacts suitable for scalable HPC data-reduction workflows.
Stand Out From the Crowd
Upload your resume and get instant feedback on how well it matches this job.
Job Type
Full-time
Career Level
Intern
Education Level
No Education Listed