GAVEL is a framework for verifying and repairing long-horizon LLM planning through an explicit graph world model. This model predicts the consequences of LLM-generated actions, detects violations, and repairs errors by following the world model. GAVEL improves single-task success by 50.6% and multi-task success by 72.7% compared to baseline LLMs.

Researchers propose a novel framework for improving interpretability of spreadsheets in LLM-driven RAG systems by splitting them into interpretable chunks using cell role annotation. This approach aims to address the 'spreadsheet-to-LLM bottleneck' by developing dimensionality-reduction techniques to flatten 2D unstructured spreadsheets into 1D text.

A new training-free agentic retrieval framework called TRACE is introduced for accountable source discovery in digital archives. TRACE outperforms existing RAG baselines and is economically feasible for heritage institutions and companies without local GPU infrastructure.

Researchers propose the End-to-End Dual Dynamic (ED$^2$) recommender, a new LLM-based sequential recommender system that integrates index generation and sequential recommendation into a unified pipeline, leveraging user-related information to improve performance.

A study on repairing stale KV caches in retrieval-augmented generation and agentic systems after document edits, proposing a budgeted recomputation approach that outperforms other repair methods.