Structure Shapes the Future of DataxLLM Systems: Retrieval, Structuring, and Reasoning

🔍 Tutorial Overview

Large Language Models (LLMs) are increasingly capable, but strong performance on knowledge-intensive and long-horizon tasks still depends on how external information is retrieved, structured, reasoned over, and maintained over time.

This tutorial presents a unified view of structure-enhanced Data × LLM systems, starting from Retrieval-Augmented Generation (RAG) and the broader Retrieving–Structuring–Reasoning (RAS) framework, then extending to structured reasoning and LLM agents.

We cover recent advances in structure-enhanced retrieval, LLM reasoning over graphs and other structured representations, agentic search, agent memory, and harness technologies for long-horizon tasks.

🔧 What You’ll Learn

[Intro Slides]

Retrieving: Retrieval and RAG with Structures

[Slides]

Explore semantic indexing, taxonomies, knowledge graphs, query generation, reranking, and structure-aware RAG. Representative topics include GraphRAG, PairSem, DynamicRAG, Hypercube-RAG, and Structure-R1.

Reasoning: Reasoning with Structures

[Slides]

Learn why flat context and free-form reasoning can fail, and how graphs, schemas, query-time structures, and structured intermediate states can improve multi-step reasoning. Representative works include SARG, StructRAG, Structure-R1, KARE, SLIDERS, and Harness-1.

Agents: Tools, Search, Memory, and Harnesses

[Slides]

Follow the evolution from tool-augmented LLMs to search agents and long-horizon autonomous systems. Topics include ReAct, FLARE, DeepRetrieval, Search-R1, s3, Harness-1, agent memory, reusable skills, and self-evolving agents.

🎯 Who Should Attend?

Researchers, practitioners, and students working on Information Retrieval, RAG, Knowledge Graphs, LLM Reasoning, Agentic Search, LLM Agents, Agent Memory, and knowledge-intensive AI applications.

⏰ Schedule

Retrieving: Enhancing the Power of Retrieval with Structures

Reasoning: Enhancing LLM Reasoning with Structures

Agents: Enhancing LLM Agents with Tools and Search

Whether you are building RAG systems, structured reasoning methods, or long-horizon agents, this tutorial highlights a common theme: structure is becoming a core component of how modern LLM systems retrieve, reason, remember, and act.