Eluna Automates Warehouse Operations with LLM Agents and SOP Compliance.

A new agentic LLM system named Eluna has been developed and deployed to automate intricate warehouse operations. This system addresses the challenge of ensuring procedural compliance and managing context overload, common issues with existing LLM agents in such environments. Eluna structures Standard Operating Procedures (SOPs) as directed acyclic graphs, enabling progressive disclosure of information and delegating tasks to parallel sub-agents equipped with persistent code execution and live data access. To meet stringent production requirements for latency and accuracy, Eluna employs an innovative training method called asymmetric episodic distillation. A powerful teacher model is refined using error memories, and then a smaller student model is fine-tuned on these corrected trajectories, internalizing improvements without incurring high inference costs. This approach has demonstrated superior performance, matching or surpassing larger off-the-shelf LLM baselines and achieving high agreement with expert human performance on real-world applications.