Inkly: A Data-Informed and Context-Aware System for HPC Job Execution
Academic Article
Overview
Overview
Abstract
High-performance computing (HPC) systems are difficult to use due to complex job scheduling, resource selection, and limited feedback on job failures. This paper presents Inkly, a data-driven HPC assistant with job intelligence that augments user workflows with insights derived from historical Slurm job data. Inkly ingests job records via sacct, stores them in a SQLite database, and computes aggregate metrics such as partition success rates, CPU and memory usage patterns, and failure distributions. These metrics are added to the prompts to guide users toward more effective job configurations. The system enforces safety through prompt filtering, command guardrails, and containerized execution using Apptainer.
