•Korea Human Resource Development Institute for Health & Welfare · Samsung Medical Center

Samsung Medical Center · KOHI Medical-AI Workforce Training, Hands-on 1: Organizing Research Material and Knowledge Management

The first hands-on session after the July theory course. On their own laptops, participants took one research topic through literature search, a Zotero library, Obsidian notes, and an Introduction draft in three hours.

AIVibe CodingClaudeHands-onmedsci-skillsWorkforce trainingKOHI

Overview

  • Date: Wednesday, 19 August 2026, 18:30–21:30 (3 hours)
  • Venue: Hippocrates Hall, 9th floor, Building B, Samsung Medical Center Ilwon Campus
  • Host: Ministry of Health and Welfare · KOHI, "2026 Medical-AI Workforce Training for Health Professionals," Samsung Medical Center track
  • Course: Organizing research material and knowledge management with AI (hands-on module 1, part 1)
  • Format: Hands-on on participants' own laptops, using published papers and practice material

What participants built

The session aimed at three outputs: a library with papers and PDFs in one place, an Introduction draft grounded in the collected literature, and an LLM-wiki of linked notes. The goal was to go through the whole path once, without needing to finish everything.

For those who missed the July theory session, I opened with a short recap: chatbots versus agents, Tool Use, the Code tab in Claude Desktop, installing and calling medsci-skills, Plan mode, and permission modes. I also set the rule early that no patient information would be uploaded; we worked only with published papers and practice files.

Five steps

  1. From a research question to literature. Participants wrote their question in one sentence and had the AI search. The first checkpoint was catching invented papers by matching against PubMed and CrossRef.
  2. Zotero as the record. We installed Zotero, the browser connector, and Better BibTeX, and connected Zotero to Claude through MCP. We kept PDFs in a local folder and linked them by path instead of uploading them to Zotero's cloud, and checked that each downloaded PDF was the right paper and that the attachment file actually existed.
  3. Notes and links. One Obsidian vault, one note per paper, then shared concepts that tie the papers together into a wiki.
  4. Introduction and references. An Introduction draft based on the organized literature, suggested citation points that participants chose themselves, and a reference list generated by Zotero.
  5. Final check. We separated what a machine can match (does the DOI exist, are authors and year correct) from what a person has to read (does each citation support the claim it is attached to).

Pitfalls we looked at

In the search example, one of the top 12 results was a retracted paper. We also covered how auto-downloaded PDFs can include the wrong paper, and how Zotero can list a PDF that is not actually there. I introduced the Remote Control feature and said I keep it off, because I would end up checking it during clinical work and family time.

Looking ahead

Part 2 on 24 August moved on to practice data: tables and figures, then Results and Discussion. As homework I suggested attaching the remaining PDFs and growing the notes to five. Thanks to the organizers and to everyone who spent the evening on this.

Host a Lecture at Your Institution

Open to invitations from hospitals, academic societies and departments. Past audiences include radiology, surgery and general clinical staff.