As researchers and developers, we often rely on tools like Zotero or Mendeley to manage our citations. They’re great for collecting and organizing references, but I’ve found them limited when it comes to searching and extracting insights from saved citations.

Over time, I’ve developed a more flexible workflow that combines traditional reference management with coding agents (like Cursor or Copilot) to make literature organization smarter and more automated.

My Workflow

Here’s how I set it up:

I create a single .bib file to store all references.

Each entry includes the paper’s download link, abstract, keywords, and notes—usually a short summary or reason for citation.

All downloaded papers are stored in a separate folder, ideally organized by topic.

For quick exploration, I use coding agents (such as Cursor or GitHub Copilot) to search and extract information directly from the .bib file.

This works especially well if the .bib file contains abstracts, keywords, and short notes—even without full-text access.

For deeper retrieval, I extend the approach using Retrieval-Augmented Generation (RAG).

This allows the coding agent to access both the .bib file and the full-text papers I’ve downloaded. Essentially, the agent “reads” from my personal library instead of querying the open web.

When prompting, I always instruct the agent to rely only on the saved citations (and downloaded papers, if available).

The response must include source references. This minimizes AI hallucination and ensures outputs are grounded in verified material.

Why This Works

This approach gives me:

• Full control over citation context and metadata
• Local, reproducible research environment
• Powerful semantic search using AI tools
• Less risk of fabricated or irrelevant information

If I only need to summarize main ideas across papers, storing abstracts in the .bib file is enough.

When I need to refer to conclusions or methodology in detail, I make sure the full paper is downloaded and indexed.

This workflow blends classic academic management with modern AI-driven retrieval—perfect for researchers who want to integrate intelligent automation without losing academic rigor.