AID2E Technical Meeting

Date: June 11, 2026

Announcements

• DOE approved the project progress report.
• James has started his position at William & Mary.

Phase I

Derek

• PR #47 (fixes test failures) - ✅ Reviewed and merged.

• PR #51 (low-level Python EIC interface)

  • Ready and reviewed.
  • Planned to merge.

• PR #53 (high-level YAML EIC interface)

  • Ready and reviewed.
  • Planned to merge.

• Identified that the CLI is not fully implemented.

• Discussing CLI implementation with Cynthia.
• Will open a GitHub issue describing the remaining implementation steps.

Cynthia

• Added support for building ePIC once per study instead of once per trial to improve execution speed.
• Working on implementing sub-job support within simulation/reconstruction workflows (e.g., π/K simulations over multiple kinematic points).
• Investigating the best way to integrate sub-job handling into the AID2E framework.

Discussion

Framework Development

• Identify the remaining work required before running production studies and generating plots for the paper.
• Begin stress-testing framework functionalities that are already implemented.
• Continue development of the EIC interface.

Multi-stage / Sub-job Workflows

• Existing PanDA implementation may provide a model for handling SLURM sub-jobs.
• Maintain a common workflow philosophy across all sub-detector implementations.
• Multi-stage workflows are critical for EIC production workflows.
• DTLZ benchmark studies can proceed independently while multi-stage support is being developed.
• Discussed overlap policies for failed trials (drich_utils.py -> apply_overlap_policy).

Action Items

1) (Amit + Derek) CLI(s) needed for DTLZ tests with different schedulers and optimizers

2) (Cynthia, anybody else?) no way to collect objectives (missing utility to map that output back to Ax/optimizer) - [drich_utils.py: collect_objectives, failed_trials_from_stage_result] 3) (Cynthia) try to figure out best way to implement per trial subjob support → merge in AID2E framework [drich_trials.py] 4) (Cynthia) PR for review after PR#53 merge for adding the no-per-trial build geometry option [drich_utils.py, prepare_trial_geometry] 5) Work on multi-stage (different schedulers) to support EIC workflows (Derek) continue work on this in connection to point 1) (Amit) Work supporting multi-stage for PandA Runner — see point 1) (Cynthia) — see number 2)
6) Converging on one single sub-detector workflow (either dRICH, BIC, far-forward) and potentially duplicate that for stress-tests — specific to subdetector workflows

* (Baptiste) — ✅ independent optimization study achieved already; how can this be included in the framework? ifarm/SLURM was done; some preliminary test with PanDA, issue with EIC environment

    *   (BNL?) help with PanDA implementation? — will discuss this offline

* (Cynthia) dRICH – check and complete; drich-example, complete implementation using AxOptimizer, SlurmScheduler/JobLibScheduler, EpicStack and configs read from workflow yaml with some additional utilities that can be merged into framework

* (Derek) BIC – ✅ The workflow introduced in PR#53 is sufficient for the paper: it’s a SOBO on the BIC AstroPix staves using the photon phi resolution as an objective at 3 kinematic points. I’ve tested the analysis script and kinematic points in the “standalone” (until we get the FW running) BIC-MOBO repo, and Amit and I have confirmed it runs as expected with Ax + SLURM/PanDA

Overleaf

• Begin drafting the paper, create an outline, distribute writing responsibilities among collaborators.

Tasks for Next Technical Meeting

Cris + Derek (+ contributors)

• Create the overall paper structure.
• Define sections and subsection outlines.

Cris

• Draft the Introduction.

Cris (+ contributors)

• Add figure templates from previous studies.
• Develop an initial narrative based on existing optimization results.

Phase II

AID2E Agent Integration

• Reviewed the preliminary AID2E agent presented on June 11.

• Successfully demonstrated:

  • Docker container builds and installs the AID2E framework.
  • Deployment of the AID2E agent.
  • Integration with the PanDA MCP framework.
  • Successful execution of a simple AID2E CLI workflow.

• Integrate the PanDA job-monitoring MCP server developed by the BNL ePIC streaming workflow team.

• Amit will continue implementation based on the presented roadmap.
• Continue roadmap planning and task distribution.