logseq/pages/PROJECTS___MOSAICO___WP2.md

Meetings

- **Meeting [[04-07-2025]]**
	- type::  [[meeting]]
	  external-links::  
	  tags:: 
	  people:: 
	  date:: [[04-07-2025]] - 11:44
	  duration::
	-

• Deliverable D2.1

  • What we're envisioning is that the protocol exposes the information about the agents in a given agent server, which then can be easily consumed by your repository (for indexing). This is the same approach as A2A agent cards and as MCP tool descriptions. A single agent server only lists its own agents: the MOSAICO repository is the one that actually indexes all the agents available across all the servers.
  • I think this is needed, because I cannot imagine people going to a web interface and filling in form field after form field with all this information. A2A precisely does this so that they could do a type of web crawler that would go through "well-known URIs" (http://host/.well-known/agent.json) to find agents at will. In fact, I suspect your repository may need to do exactly this.
  • ✉: Web Link

• MENTORING/TESISTI

  collapsed:: true
  • Marco Giarrusso
  • Mariama Celi S. de Oliveira
    • Rehearsal presentation
      • what about adding an explanatory workflow to show an orchestration of agents with Dify or AutoGPT?
    • Collaborative optimization task--> explain
    • Just motivate the choices for the configurations slide 8
    • Slide 9
      • too small
    • Slide 13
      • What do you mean with Each pipeline runs only once?
  • Motunrayo Ibiyo
    • Rehearsal presentation collapsed:: true
      • Mention MOSAICO at the beginning where we are studying the usage of LLM-based multi agent systems to support software engineering tasks
      • Very fast, reduce the content and speak slowly. Too many slides!!! You should cut half of them!!!
      • It's easy to get lost in the details
      • Slides 4
        • What are the limitations of existing approaches?
      • Slide 6
        • with-out (typo)
      • Slide 7
        • Put the updated reference
      • Slide 8
        • highlight the different components while presenting (one by one)
      • Slide 10
        • 
        • what does it mean?
      • Slide 11
        • Many typos
          • a file in the package is...
      • Slide 12
        • 
          • too many things
      • Slide 14
        • 
        • It's too much,
      • It's more beneficial presenting some explanatory examples instead of all these numbers, in line with what you have with slide 16.
  • TOOLS evaluated
    • Qualitative + Quantitative
      • AutoGen
      • AutoGPT
      • Dify
      • SemanticKernel
      • LlamaIndex
      • Heystack
    • Qualitative
      • Those from Motunrayoi