Coach Handbook
Everything an instructor needs to run the AgentOps Control Tower RVAS. Read this first, then review the per-challenge coach guides.
🧑🏫 Coaching philosophy: facilitate, don't solve. Let teams struggle productively. Ask leading questions, point to the right doc, and reveal the reference solution only when a team is genuinely stuck or running out of time. The
resources/directory is the full answer key — you decide how much to expose and when.
Who this is for
- Audience: platform/cloud engineers, data engineers, FinOps practitioners, and AI engineers at the customer. Level 200–400.
- Team size: 3–5 people. Aim for a mix of skills per team so each challenge has an owner.
- Coach ratio: 1 coach per 1–3 teams works well.
What teams build
A Fabric-based control tower that ingests and correlates telemetry from Azure services and
Foundry-built agents, delivering dashboards and alerts for reliability, cost, and performance.
See docs/architecture.md. The reference implementation lives in
resources/.
Sample agenda (2 days)
| Time | Day 1 | Day 2 |
|---|---|---|
| Morning | Welcome + mission briefing (30 min) · Challenge 0 · Challenge 1 | Recap · Challenge 4 |
| Afternoon | Challenge 2 · start Challenge 3 | Challenge 5 · Challenge 6 (stretch) · Showcase + retro |
Compressed to 1.5 days? Pre-provision Challenge 0 prerequisites, run Challenges 1 & 2 in parallel (split teams), and treat Challenge 6 as optional.
⏱️ The biggest schedule risk is environment access (roles, quota, Fabric enablement). Drive the prerequisites before Day 1. Have a backup subscription and a pre-started Fabric trial in your pocket.
How to run each challenge
- Frame it (5–10 min). Set the scene with the challenge's mission hook and objectives. Use the "Talking points" in each coach guide as mini-lecture material.
- Let them work. Teams self-organize. Circulate, unblock, ask questions. Resist hand-holding.
- Checkpoint (✅). Validate against the success criteria using the "Checkpoint verification" in the coach guide. Award points if you're scoring.
- Debrief (5 min). Reinforce why it mattered before moving on.
Reveal policy for resources/
| Team is… | Do this |
|---|---|
| Cruising | Keep the reference hidden; push them toward stretch goals and "explain it to me" |
| Steady | Point to the relevant resource README, not the solution commands |
| Stuck (15+ min) | Share the specific command/snippet from the coach guide |
| Out of time | Have them deploy the reference asset directly so they can continue the arc |
The arc matters more than any single challenge — don't let a team fall off the path. It's fine to fast-forward a team through Challenges 1–2 (Azure foundation) so they reach the Fabric core (3–6), which is the point of the event.
Per-challenge guides
| # | Guide | Focus |
|---|---|---|
| 0 | challenge-00-guide.md | Access, tooling, Fabric capacity |
| 1 | challenge-01-guide.md | Deploy + verify agent telemetry |
| 2 | challenge-02-guide.md | ADLS Gen2 landing zone |
| 3 | challenge-03-guide.md | Lakehouse, shortcuts, Mirroring |
| 4 | challenge-04-guide.md | Medallion pipeline |
| 5 | challenge-05-guide.md | Semantic model + dashboards |
| 6 | challenge-06-guide.md | Alerts, chargeback, RLS |
Logistics
- Shared Fabric capacity. One trial/paid capacity can back every team. As capacity admin, assign each team's workspace to it. Give each team its own workspace.
- Naming. Agree a convention up front, e.g.
rg-agentops-<team>and workspaceAgentOps-<team>, so resources are easy to find and clean up. - Cost. Low by design (see FAQ). Use
gpt-4o-mini. Pause paid capacity overnight. - Cleanup (end of event). In each workload dir:
azd down --purge. Delete Fabric workspaces and, if you started a trial, let it expire (or end it). Remove cost exports.
Scoring (optional)
The rubric lives in challenges/README.md. For a closing
showcase, have each team demo their Control Tower answering the three business questions
(reliability, cost, performance). Judge on correlation and usefulness, not pixel polish.
Facilitation tips
- Keep the three business questions on a whiteboard all event. Every challenge ladders up to them.
- Celebrate the first end-to-end trace (Challenge 1) and the first Direct Lake visual (Challenge 5) — they're the "aha" moments.
- If energy dips in the data-engineering middle (Challenge 4), remind teams the payoff is the dashboard two hours away.