An Agentic Scholarly Workflow
Gone are the days where Large Language Models were simply confident, but supremely untrustworthy, chat bots that provided you with perfectly title articles by famous scholars that you then found out did not actually exist. Now with coding harnesses like OpenCode, Hermes, or Claude Code, Model Context Protocols, and agentic skills written by experts from around the world, LLM’s are turning into capable research tools that can check their own answers, help researchers dive deeper than ever into their data, and even automate some aspects of the research process. In this workshop we will explore how to deploy both locally hosted and cloud-based LLM agents using a coding harness, connect that harness to scholarly MCP servers, find research based agentic skills, and tie them altogether into a workflow that you can carry with you from machine to machine.
By the end of this session participants will be able to:
- Connect both local and cloud LLM models to coding harnesses
- Use MCP servers to mitigate LLM hallucinations
- Understand what agentic skills are and how to both find existing ones and create their own
- Deploy their own scholarly agent using a containerized workflow
Workshop Preparation
Participants are required to have a laptop or other computing machine capable of utilizing a command line interface. For Windows users this will mean utilizing Windows Subsystem for Linux, which should be installed before the session. It is also required that participants have Docker, either desktop or command line, installed on their systems.
Facilitator Bio
John Fink is the Digital Scholarship Librarian in the Sherman Centre.
Workshop Slides
Coming soon.