LLM-Agents-Papers

Implements a multi-level hierarchical classification system that organizes LLM agent research papers into primary categories (Survey, Technique For Enhancement, Interaction Paradigms, Application Domains) with subcategories, enabling structured navigation of a rapidly evolving research landscape. The system uses a README.md-driven taxonomy definition that maps papers into logical groupings by research methodology, application domain, and temporal evolution, making it easier for researchers to discover papers aligned with specific research interests without manual filtering.

Maintains versioned paper metadata organized by publication year (parsed_v5 directory with JSON files per year) and tracks research focus evolution across 2023, 2024, and 2025, allowing researchers to identify which techniques, paradigms, and applications gained prominence in specific years. The system uses a time-series approach where papers are indexed by year and linked to their corresponding research focus areas, enabling analysis of how LLM agent research priorities have shifted over time and which emerging areas are gaining traction.

Enables filtering papers by enhancement technique categories (e.g., prompt engineering, chain-of-thought, retrieval-augmented generation, tool use, planning, memory mechanisms) by mapping papers to specific methodological approaches used to improve LLM agent capabilities. The system uses a technique-centric organization where papers are indexed by the enhancement methods they propose or evaluate, allowing researchers to find all papers related to a specific improvement strategy regardless of application domain or interaction paradigm.

Classifies and organizes papers by interaction paradigm categories (e.g., single-agent, multi-agent, human-in-the-loop, tool-mediated interaction) to enable researchers to find papers addressing specific agent interaction models and communication patterns. The system uses a paradigm-centric dimension where papers are indexed by the type of agent interactions they address, allowing discovery of papers relevant to specific architectural interaction patterns independent of the enhancement techniques or application domains involved.

Organizes papers by application domain categories (e.g., game agents, autonomous systems, code generation, question answering, robotics) to enable researchers to find papers addressing specific real-world use cases and domain applications of LLM agents. The system uses a domain-centric indexing approach where papers are mapped to their primary application context, allowing discovery of domain-specific agent implementations, benchmarks, and evaluation methodologies.

Provides dedicated organization and curation of papers specifically focused on multi-agent systems, including agent coordination, communication protocols, emergent behaviors, and collaborative problem-solving. The system uses a specialized subcategory within the broader taxonomy to collect papers addressing multi-agent architectures, enabling researchers to focus on papers dealing with agent-to-agent interactions and collective intelligence rather than single-agent systems.

Provides a download_pdf.py utility script that automates bulk downloading of research papers from URLs stored in papers_v5.json metadata, enabling researchers to build a local paper collection without manual URL processing. The script uses paper metadata to construct download requests and manage file organization, allowing researchers to create an offline research library indexed by the repository's taxonomy for local searching and analysis.

Maintains multiple versions of paper metadata (parsed_v4, parsed_v5 directories) with version-specific JSON schemas, enabling schema evolution and backward compatibility as the repository's data model changes. The system uses a versioning approach where each metadata version is stored separately, allowing researchers to access papers using different schema versions and supporting gradual migration to newer metadata formats without breaking existing workflows.

Identifies and organizes survey papers that provide comprehensive reviews and analyses of the LLM-based agents field, offering summaries of current research, methodologies, challenges, and future directions. The system uses a dedicated 'Survey' category in the taxonomy to collect meta-level papers that synthesize research across multiple techniques, paradigms, and domains, enabling researchers to quickly access high-level overviews of the field rather than reading individual technique papers.

Provides a contributing guide that documents the process for community members to submit new papers, propose taxonomy changes, and improve the repository's organization. The system enables collaborative curation where researchers can contribute papers, suggest new categories, and refine classifications, maintaining the repository as a living resource that evolves with the research community rather than a static snapshot.