academic-literature-mining-skill

This Skill addresses core pain points in academic literature management: scattered literature, incomplete citations, and cross-contamination of data. In practical research, scholars often need to manually collect papers, verify metadata, and organize citations—a process that is both tedious and error-prone. This tool automates the entire workflow, from search and filtering to indexing and storage, ensuring authoritative and complete citation records for each paper while strictly isolating data between different papers.

Usage is straightforward: simply describe your research topic in natural language, such as "building a corpus for large language model evaluation." The tool will automatically invoke the Rust CLI litmine to search academic databases, parse identifiers (DOI/arXiv/OpenAlex), download PDFs, and extract complete citation metadata. It then uses NVIDIA Nemotron Embed/Rerank models to generate embeddings, stores them in a Qdrant vector database, and supports retrieval by semantics or keywords. The entire process eliminates the need for manual file or database organization.

It is particularly suitable for researchers building reproducible academic corpora, scholars writing literature reviews, and lab teams managing large volumes of papers. Users already familiar with command-line tools who need to efficiently process extensive literature will see significant improvements in mining and management efficiency. It also fits technical research scenarios requiring comparison of different model performances.

It is recommended to use an independent Compose project for each new paper to isolate data and avoid vector database contamination. Pre-configuration of the NVIDIA Nemotron API (or a local GPU model) and a Qdrant container is required. Note that the tool depends on a Rust environment and Docker Compose, and is recommended for Linux or macOS. Additionally, each paper must have a parseable persistent identifier (e.g., DOI/arXiv); it is not suitable for informal literature lacking such identifiers.