Warehouse operations are governed by Standard Operating Procedures (SOPs) that encapsulate complex, multi-system decision logic, which must be executed reliably under strict time constraints. However, existing LLM agents lack mechanisms to enforce procedural compliance and degrade under the context overload introduced by full SOP specifications. We present Eluna, a production-deployed agentic system for reliable SOP execution.
Eluna is a graph-guided, multi-agent framework that encodes SOPs as directed acyclic graphs with progressive disclosure, delegating independent tasks to parallel sub-agents, each equipped with persistent code execution and live data access. To meet production latency and accuracy needs, we utilize asymmetric episodic distillation where a strong teacher is improved through episodic error memories, followed by fine-tuning a smaller student on the corrected trajectories with memory stripped, allowing for internalized corrections without inference-time overhead.
In a benchmark of 13 tasks and two production applications, our fine-tuned models match or exceed their teacher, outperform all larger off-the-shelf baselines, and achieve 94% expert agreement on the ticket processing application.
Blogger's Review: Eluna effectively addresses the complexity and real-time demands of warehouse operations through its graph-guided multi-agent framework. Its asymmetric episodic distillation method enhances model performance while reducing inference burden, showcasing a significant breakthrough in the industrial application of LLMs.