Orchestrated Agentic Experimentation
Multimedia Hall
Faculty of Electrical Engineering, University of Ljubljana
Course provider: Faculty of Electrical Engineering, University of Ljubljana
Instructors: Assoc. Prof. Dr. Janez Perš, Dr. Jon Muhovič
Learning objectives: Participants will learn the principles of Orchestrated Agentic Experimentation: a disciplined, human-governed workflow in which a scientist or researcher uses a strategic AI model and an autonomous execution agent to move from a research question toward a reproducible experimental setup and preliminary evidence.
Content: The full-day workshop, Orchestrated Agentic Experimentation, will run from 8:30 to 16:00, with a coffee break and lunch in between. Participants will learn how to transform a research idea into an experimental brief, how to efficently obtain related work, datasets and code, and use a reasoning model to help with using datasets, models, codebases and evaluation protocols. They will learn how to prepare a controlled workspace with access to data and compute, and how to delegate dataset preparation, model download, environment setup, code installation and preliminary experimental runs to an autonomous agent.
Learning outcomes: Participants will be able to prepare an actionable experiment brief, define roles between human scientist, strategic model and execution agent, create a research-oriented project constitution, run or stage a small reproducible experiment, and assess whether the resulting evidence supports a scientific claim. The workshop is intended for researchers, scientists, PhD students, research software engineers and AI/ML practitioners.
Special requirements: Participants should bring their own laptop and have access to a paid ChatGPT Plus/Pro account or an equivalent environment (Claude with Claude Code). Familiarity with ChatGPT/Claude is required, familiarity with agentic coding is a bonus, but not required. Basic familiarity with Python, Git, the terminal and scientific computing is recommended. Knowledge of scientific experimentation process for participant's field is very beneficial.
For the practical part, we will use a separate controlled workspace, such as WSL2, a Linux virtual machine, a remote GPU machine or a prepared SLAIF/HPC environment. Participants who wish to work on their own topic should bring a short research question and only public, anonymized, synthetic or otherwise shareable sample data.