Large Language Models (LLMs) such as ChatGPT have significant limitations that hinder their full utility in contexts where traceability and reliability of responses are required.
AI Agents help reduce or eliminate these limitations by employing patterns that mimic human organization and activities, and by structuring the memory and "context" provided to LLMs.
This course is intended for advanced users and developers who wish to deepen their knowledge of techniques to refine Generative AI outputs and create autonomous workflows. autonomi.
The following topics will be covered:
Introduction to HPC and LLMs
Utility of AI Agents
Implementing a framework "from scratch"
LangGraph
Advanced functionalities (tool-state interaction, interrupts, commands)
Multi-agent systems
MCP (Model Context Protocol) Servers
RAG (Retrieval-Augmented Generation)
Two seminars on use cases in astronomy
Proficiency in Python programming, experience with LLMs, and knowledge of prompting (Note: the course does not cover prompt engineering in depth). The course will be tought in Italian, with the exception of one seminar which will be held in English.
The final day will be dedicated to a discussion on the prospects of Generative AI and Agentic AI within INAF.
The course consists of lectures, project work, and development activities; participants are encouraged to propose real-world problems and use cases to be developed during the practical sessions.
PARTICIPATION AND COSTS
In-person: Attendance (which is essential for hands-on activities and project work) is limited to 30 people.Remote participation is available but it is limited to lectures.
NB: A tutti i partecipanti (sia in presenza che da remoto) è richiesta la registrazione al sito indico dell’evento:
Materials and Recordings
Video Recordings of the lectures are now available at the following links (INAF @inaf.it account is required for access):
Villaescusa seminar 1a parte Part 1 – Part 2
Scripts and slides are available in this repository:
https://gitlab.hpc.cineca.it/rmioli00/agents_101
The course is free of charge. Participants must cover their own travel and mission expenses through their own funds.
STEERING AND LOCAL ORGANIZING COMMITTEE
This course is organized by the GenIA study group of the USC-C (*) in collaboration with Cineca / AI Factory. gruppo di studio GenIA dell’USC-C (*) assieme a Cineca / AI Factory.
We thank OAS for their logistical and financial support, and the USC-C for their financial support.
Giulio Capasso (OACN)
Andrea Bulgarelli (OAS)
Luca Castaldini (OAS)
Nicolò Parmiggiani (OAS)
Simone Riggi (OACT)