AgentOps Workshop — author guide

Delivering the workshop? Use the instructor site instead: https://studious-couscous-1qj2kwp.pages.github.io/ The PowerPoint deck and narrated video are one click away from the home page.

This README is for authors and builders working on the workshop content itself. The repository has two audiences served from the same source:

Audience	Entry point	Purpose
Instructor delivering the workshop	GitHub Pages site	Download slides, video, speaker script.
Author building the workshop	This README + prep/	Edit content, regenerate artefacts, publish.

Tracks

Track	Status	Source
AgentOps Briefing (~1 hour)	Ready to deliver	short/
AgentOps Value Delivery Workshop - Value-Based Delivery (VBD) (~8 hours)	Facilitator-ready design-lab plan	long/

Repository layout

Path	Audience	What's inside
short/	Instructor	Agenda, run-of-show, speaker script, slides.md, slides.pptx, narrated video (released), images/.
long/	Instructor	AgentOps VBD Workshop agenda, lab roadmap, observability strategy, and facilitator-ready design-lab pages.
instructor/	Instructor	Cross-track delivery notes.
assets/	Instructor	Shared images, slide exports, sample data.
index.md, _config.yml, Gemfile	Instructor (Jekyll site)	GitHub Pages source.
prep/	Author	References, build tools, planning notes, intermediate audio/video. Excluded from the published site via _config.yml. See prep/README.md.
.github/copilot-instructions.md	Author	Repository-specific guidance for GitHub Copilot.
README.md (this file)	Author	You are here.

Authoring workflow

AgentOps Briefing is built from three master sources:

1. short/slides.md (Marp) → short/slides.pptx
2. short/speaker-script.md (verbatim narration with **SP1:** / **SP2:** turns)
3. short/images/ (diagrams generated by prep/tools/make_diagrams.py)

From those, the build pipeline produces the narrated video. See prep/README.md for the exact commands.

Regenerating the narrated video

az login --tenant 16b3c013-d300-468d-ac64-7eda0820b6d3

python prep\tools\render_speech.py        # SSML synthesis -> prep\short\audio\*.wav
python prep\tools\render_slides_hd.py     # PPTX -> prep\short\video\*.png (1920x1080)
python prep\tools\build_video.py          # ffmpeg assembly -> short\agentops-short-video.mp4

Publishing the video

The narrated MP4 is committed directly to short/agentops-short-video.mp4 and served as a static asset by GitHub Pages. The Pages site is private (only Azure org members can reach it), so embedding it same-origin is what lets the browser play it - cross-origin <video> requests to the private GitHub Release URL return 404 and produce an empty player.

To keep the file under GitHub's 100 MB per-file push limit, the build pipeline finishes by re-encoding the audio (ffmpeg -c:v copy -c:a aac -b:a 48k -ac 1 -movflags +faststart). Today's output is ~85 MB for ~46 min at 1080p.

If a future video genuinely cannot be compressed below 100 MB, switch to GitHub Releases AND mirror the asset somewhere the browser can reach without auth (e.g., a public Azure Blob with anonymous read, or an unlisted YouTube embed). The release URL by itself will not play in the embedded <video> while the repo is private.

Local preview of the Pages site

bundle install
bundle exec jekyll serve

Then open the local URL shown by Jekyll. The prep/ folder, .venv/, and vendor/ are excluded from the build.

Publishing checklist

Before making any release or significant content change visible:

• Confirm baseurl in _config.yml matches the repository name.
• Review screenshots, traces, links, and sample data for sensitive information.
• Rebuild the video if the speaker script or slides changed.
• Publish a new GitHub Release for the video and update the Pages link if needed.

Main storyline

1. Agents are moving from prototype to production.
2. Production needs evidence: evals, gates, observability, safety, and operations.
3. AgentOps provides the operating model.
4. Foundry remains the control plane.
5. AgentOps practices create the repo and CI/CD release-readiness loop.
6. Observability connects runtime behavior, traces, incidents, and production learnings back to evaluation.
7. The demo shows a regression caught before release, fixed, observed, and promoted with evidence.