Shifu: A self-learning framework for automating root cause analysis in logistics operations

Modern logistics networks face a critical challenge in performance documentation that consumes substantial resources yet suffers from inconsistent quality, limited expert review, and context-specificity. We present Shifu, an adaptive knowledge acquisition system for automated root cause analysis that learns continuously from operational feedback without requiring gold standard examples. Shifu integrates targeted machine learning, agent-based data analysis, utility-driven insight prioritization, and active learning through a comprehensive feedback loop. We evaluated Shifu across five North American logistics facilities over a two-week deployment, demonstrating improvements in content quality (reaching 87.9% acceptance within one week), effective feedback incorporation (89.5% closure rate), and knowledge expansion (44% metric growth in key categories). Our results show a 4X improvement over baseline systems, with Shifu self-adapting to facility-specific operational contexts while continuously enhancing its analytical capabilities. This approach transforms resource-intensive analytical processes by complementing rather than replacing human expertise, providing a blueprint for continuous learning systems in domains with subjective quality criteria, specialized operational contexts, and limited supervision.