Coach Hub
Your job is to create momentum, not to become the team’s keyboard. In What The Hack, great coaching means asking better questions rather than giving faster answers.
Coaches for this event should focus on facilitation, pacing, and unblock strategy. Student-facing challenge pages stay public; full solution guides remain in the repository so you can use them selectively when a team is truly stuck.
Event Day Checklist
Pre-event
- Verify participants have a working GitHub path and an Azure subscription route
- Confirm you can access the repo locally, including
solution.mdfiles in each challenge folder - Skim the challenge sequence so you know where setup ends and AI work begins
- Check the event’s escalation path for Azure subscription or portal issues
During the event
- Start every team in Foundations (Step 1) and confirm their environment is actually usable
- Watch for drift: teams often think they are blocked by code when they are really blocked by setup
- Keep teams time-boxed and encourage strategic skipping if the event clock gets tight
- Use questions first, direct fixes second
Post-event
- Help teams capture what they built and what they learned
- Encourage cleanup of unused Azure resources if the event requires it
- Share next-step learning resources for teams that want to keep building
Facilitation Principles
Ask, don’t tell
Good prompts for teams:
- “What changed right before it stopped working?”
- “Where do you think the request is failing: auth, config, or code?”
- “What does success look like for this challenge?”
- “If you had to test one assumption first, which would it be?”
Identify struggle vs. stuck
Healthy struggle looks like experimentation, note-taking, and narrowing hypotheses. A stuck team repeats the same failing action, cannot name the blocker, or loses confidence in the next step. Intervene when they stop learning from the attempt.
Celebrate progress, not just completion
Call out good debugging, clear prompt design, and smart teamwork. Teams gain energy when you recognize real progress between checkpoints.
Per-step / per-challenge timing guide
Tier 1 · Foundations (one guided challenge, four ordered steps):
| Step | Expected Time | Warning Sign | Intervention |
|---|---|---|---|
| 1 Setup & Provisioning | 30 min | >45 min | Check Azure subscription, Codespaces readiness, azd up, and Foundry access. |
| 2 Model & Playground | 45 min | >1 hr | Verify model deployment, endpoint, and keyless auth (DefaultAzureCredential). |
| 3 Your First Agent | 1 hr | Agent answers out-of-scope/unsafe asks | Review persona, system instructions, and guardrails. |
| 4 Knowledge Base (IQ) | 1.5 hr | Answers hallucinate or lack citations | Check indexing, grounding data quality, and IQ knowledge base wiring. |
Tier 2 · Advanced (modular, any order — each assumes the Foundations end-state):
| Advanced challenge | Expected Time | Warning Sign | Intervention |
|---|---|---|---|
| Action Tools | 1–1.5 hr | Tool fires without approval, or never fires | Inspect the MCP requires_action loop and approval gate. |
| Evaluation & Red Teaming | 1–1.5 hr | Teams ignore metrics they do not like | Reframe metrics as design feedback; check the CI score gate. |
| Tracing & Observability | 1 hr | No spans land in App Insights | Confirm GenAI tracing env vars are set before SDK import. |
| Deploy as a Hosted Agent | 1–1.5 hr | Image builds but endpoint 401/500s | Check ACR push, agent.yaml, and per-agent managed identity. |
Common blockers across all challenges
| Blocker | What it usually means | What to do |
|---|---|---|
| “Azure is broken” | Wrong tenant, missing subscription, or quota issue | Re-check account, subscription, and region before touching code. |
| “The model is not responding” | Deployment mismatch or bad credentials | Confirm the exact deployment name, endpoint URL, and key source. |
| “The agent answers anything” | Persona/guardrails not applied | Re-check system instructions and refusal behavior for out-of-scope asks. |
| “RAG answers are off-topic” | Weak retrieval or poor chunk quality | Verify indexing, source data, and whether grounding is actually enabled. |
| “Nothing works anymore” | Several changes landed at once | Roll back to the last known-good step and recover incrementally. |
Per-challenge coach notes
Each challenge has a coach-only reference page with the expected solution path, common blockers, and timing tips. These pages are hidden from the main navigation so students do not stumble into them.
| Challenge | Coach view |
|---|---|
| Foundations | Coach notes |
| Advanced — Action Tools | Coach notes |
| Advanced — Evaluation & Red Teaming | Coach notes |
| Advanced — Tracing & Observability | Coach notes |
| Advanced — Deploy as a Hosted Agent | Coach notes |
| Extra — Fabric IQ | Coach notes |
| Extra — Give It a Voice | Coach notes |
| Extra — Magentic Workflows | Coach notes |
| Extra — Hosted Long-Running Agents | Coach notes |
| Extra — Build a UI | Coach notes |
| Extra — Copilot-Assisted Build | Coach notes |
Solution guides are in the repo
Coach solution guides are published as themed pages on this site (see the table above) and are also available as raw files in the repository. Clone the repository and use challenges/*/solution.md as a back-pocket reference when teams need structured rescue help without opening a browser.
Use solution guides to restore momentum, not to short-circuit discovery.
Emergency resources
- Microsoft Foundry documentation: https://learn.microsoft.com/azure/foundry/
- Azure status page: https://azure.status.microsoft/en-us/status
- Portal sign-in/account issues: confirm tenant, subscription, and event instructions first
- Event escalation: follow your organizer’s support route for Azure Pass, quota, or tenant problems