Skip to content

Multi-agent RL for fusion example#166

Open
jbelof wants to merge 5 commits into
amd:mainfrom
jbelof:main
Open

Multi-agent RL for fusion example#166
jbelof wants to merge 5 commits into
amd:mainfrom
jbelof:main

Conversation

@jbelof

@jbelof jbelof commented Jun 13, 2026

Copy link
Copy Markdown

Contributes a tutorial example on how to build a multiagent system for fusion capsule design using reduced order models, autogen for agentic swarm communication and orchestration, and MCP serving examples of tooling for the physics calculation.

Jon Belof and others added 5 commits June 1, 2026 05:57
… update docs

Bring the icf_4agent example up to date with the latest working version.

Features / fixes:
- Conversational-PI debrief: after a campaign, users can chat with the Principal
  Investigator about the results ("why did the best design win?", "how robust is
  it to defects?"). Adds a tool-free conversational PI agent, a session-scoped
  store of all campaign results, multi-turn memory, and intent-based routing that
  separates analysis questions from run requests (app.py).
- Multi-round campaign fix: AutoGen's max_round counts messages, not design
  iterations, so campaigns ran only ~1 round. Convert the design-round constants
  into a message budget and wire is_termination_msg so the PI's TERMINATE /
  yield-convergence actually stops the chat early (app.py).
- Add icf_viz.py (implosion movie + diagnostic time-history plots) and the AMD
  logo assets used by the Gradio UI and media mode.
- Update README.md and docs/summary.tex to document the debrief mode and the
  four operational modes.
- Add .gitignore to exclude the venv, caches, logs, and generated media.

Note: docs/summary.pdf is NOT regenerated here (no pdflatex available in this
environment); rebuild it from summary.tex before relying on the PDF.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Regenerate the manuscript PDF (via tectonic) so it reflects the documentation
updates from the previous commit: the conversational-PI debrief subsection, the
four-operational-modes description, and the added conclusions item. The PDF was
previously stale relative to summary.tex.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Updated the diagram in the README to reflect changes in the structure.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant