Eluna: An Agentic LLM System for Automating Warehouse Operations with Reasoning and Task Execution
2026-07-09 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors present Eluna, a system designed to reliably follow complex warehouse procedures that involve many steps and tight timing. Eluna breaks down these procedures into smaller tasks handled by multiple agents working in parallel, each with its own code and real-time data access. To make the system both accurate and fast, the authors use a training method called asymmetric episodic distillation, which helps a smaller model learn from corrections made by a stronger teacher model without slowing down performance. Their tests show Eluna performs as well as or better than other models and agrees with human experts 94% of the time in one real-world task.
Standard Operating Procedures (SOPs)Large Language Models (LLMs)Directed Acyclic GraphsMulti-agent SystemsEpisodic DistillationParallel Task DelegationProduction LatencyModel Fine-tuningExpert AgreementWarehouse Operations
Authors
Ning Liu, Kalle Kujanpää, Zhaoxuan Zhu, P Aditya Sreekar, Kaiwen Liu, Chuanneng Sun, Jorge Marchena Menendez, Matthew Bales, Tianyu Yang, Shahnawaz Alam, Rose Yu, Baoyuan Liu, Kristina Klinkner, Shervin Malmasi
Abstract
Warehouse operations are governed by Standard Operating Procedures (SOPs) that encode complex, multi-system decision logic, which must be executed reliably under strict time constraints, yet LLM agents lack mechanisms to enforce procedural compliance and degrade under the context overload full SOP specifications introduce. 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 and delegates independent tasks to parallel sub-agents, each with persistent code execution and live data access. To meet production latency and accuracy needs, we use asymmetric episodic distillation where a strong teacher is improved through episodic error memories, then a smaller student is fine-tuned on the corrected trajectories with memory stripped, internalizing corrections without inference-time overhead. On a 13-task benchmark and two production applications, our fine-tuned models match or exceed their teacher, beat all larger off-the-shelf baselines, and reach 94% expert agreement on the ticket processing application.