Managing the rapidly growing scholarly corpus poses significant challenges in representation, reasoning, and efficient analysis. An ideal system should unify structured knowledge management, agentic planning, and interpretable execution to support diverse scholarly queries – from retrieval to knowledge discovery and generation – at scale. Unfortunately, existing RAG and document analytics systems fail to achieve all query types simultaneously. To this end, we propose AgenticScholar, an agentic scholarly data management system that integrates a structure-aware knowledge representation layer, an LLM-centric hybrid query planning layer, and a unified execution layer with composable operators. AgenticScholar autonomously translates natural language queries into executable DAG plans, enabling end-to-end reasoning over multi-modal scholarly data. Extensive experiments demonstrate that AgenticScholar significantly outperforms existing systems in effectiveness, efficiency, and interpretability, offering a practical foundation for future research on agentic scholarly data management.

Modern LLM-powered question answering (QA) systems are no longer merely single-step generators. They increasingly operate as analytical pipelines that interpret user intents from natural language questions, plan reasoning steps, bind them to tools or computational resources, execute workflows adaptively, and synthesize verified conclusions. Recent research has introduced a wide range of methods to improve different stages of this process, but these contributions are often studied in isolation or organized around specific application scenarios and generation paradigms. To provide a structured view of this rapidly evolving landscape, we adopt a pipeline-oriented perspective inspired by database query processing. We propose a unified five-stage abstraction: Query Rewriting, Logical Planning, Physical Planning, Execution & Adaptation, and Synthesis & Verification. We focus on QA systems over textual and tabular data, two fundamental modalities that represent unstructured and structured information processing. Within this framework, we develop a taxonomy that maps existing methods to their corresponding stages, distills stage-specific design objectives, and analyzes cross-stage interactions. By reframing LLM-powered QA as an optimizable, engineered pipeline, this survey identifies key research gaps and outlines a principled roadmap for the design of next-generation QA systems.