| name | demo-runbook |
| description | Presenter guide for walking Anna (or anyone) through the SQL + AI in Action demos live on stage. Use when the user asks to "walk me through the <demo> demo", "what's my next step / what do I say next / give me the next prompt", "how do I present <demo>", or is mid-demo and needs the next beat's narration, literal prompt, and expected result. Routes to the per-demo PRESENTER-RUNBOOK.md files. Covers zavacliniq, zavahospital, azuresqlcontainer (live-site), and hyperscale. Does NOT build/deploy — hand off environment lifecycle to the demo-specific skills. |
Demo Runbook — presenter guide
Guide the presenter (Anna) through any of the four demos live, one beat at a time:
tell her the next thing to do, the literal prompt to type, and the line to say —
then wait for her to confirm before moving on. Each demo has a self-contained presenter
runbook; this skill is the pace-setter and router over them.
The four demos and their runbooks
| If Anna means… | Demo | Presenter runbook | Build/deploy skill |
|---|
| "live site", "on-call", "incident", "mitigate", "Zava On-Call Console" | azuresqlcontainer | azuresqlcontainer/PRESENTER-RUNBOOK.md | zavalivesite-demo (lifecycle) + live-site-sql (triage agent) |
| "hyperscale", "loan scoring", "ZavaFin", "narrative search", "Act 3" | hyperscale | hyperscale/demo/PRESENTER-RUNBOOK.md | hyperscaleai |
| "zavacliniq", "outpatient", "clinic", "care plan", "Foundry Local", "laptop" | zavacliniq | sqlserver2025/zavacliniq/demo/PRESENTER-RUNBOOK.md | zavacliniq |
| "zavahospital", "inpatient", "hospital", "NIM", "AKS", "doctor notes" | zavahospital | sqlserver2025/zavahospital/demo/PRESENTER-RUNBOOK.md | zavahospital |
How to guide Anna (the loop)
- Identify the demo. If it's ambiguous, ask one short question with the four options.
If she's clearly mid-demo (e.g. "next step"), use the demo already in play.
- Read that demo's
PRESENTER-RUNBOOK.md before answering — it is the source of truth
for beats, literal prompts, talk track, and expected results. Do not invent prompts or
narration; quote them from the runbook.
- Deliver ONE beat at a time, in this shape, then stop and wait:
- Do: the physical action (open file X, run script Y, switch to browser).
- Type: the exact literal prompt/query in a fenced block, ready to copy/paste.
- Say: the one-line talk track (verbatim from the runbook).
- Audience sees: the expected result — so Anna can confirm it worked.
- Wait for confirmation ("done", "next", "it worked") before advancing. If she says
something's wrong, jump to the runbook's If it breaks section for that beat.
- Track position. Keep a light note of which beat/step she's on so "next" always advances
correctly. If she asks "where am I", restate the current beat.
Setup / reset / teardown → hand off, don't duplicate
This skill guides the walkthrough. For environment lifecycle, route to the demo's build skill
and surface the exact trigger phrase (the runbooks list them under Before you start / Reset):
- azuresqlcontainer —
prep the demo, open the live site, insert incident 5012,
run hybrid search for incident 5012, reset the agent test state,
tear down the demo → zavalivesite-demo skill. The on-stage triage prompt
Mitigate incident 5012 @live-site-sql is owned by the live-site-sql agent/skill.
- hyperscale —
build the hyperscale demo → hyperscaleai skill. Live run is
Run-LiveDemo.ps1 -ReplicaDatabase zavalending_NamedReplica -Pause. Step 1 (00-reset.sql) is the reset.
- zavacliniq —
deploy zavacliniq → zavacliniq skill (Setup-FoundryLocal.ps1 +
Deploy-ZavaClinIQ.ps1). Re-deploy to reset (deploy drops + recreates the DB).
- zavahospital —
deploy zavahospital → zavahospital skill (stand up NIM-on-AKS first,
then build/sql/ 00–14). Live scripts only read/EXEC — just re-run to reset.
Beat map at a glance (for fast routing — always confirm against the runbook)
- azuresqlcontainer (~6:30): 0 open page → 1 schema → 2 five-feature INSERT → 3 hybrid search
(plan in SSMS) → 4 agent loop (
Mitigate incident 5012 @live-site-sql) → 5 local-vs-cloud.
- hyperscale (4 steps): 1 reset → 2 Narrative Search (replica) → 3 DML on vector-indexed table
→ 4 AI Loan Scoring (Phi-4 + ledger). Optional Part 2: Foundry agent via DAB/MCP.
- zavacliniq (6 scripts, only 04 + 06 execute): 01–03 narrate → 04 vector search (~400 ms)
→ 05 narrate → 06 care plan (~60–105 s). Closer: SSMS Copilot clinician prompts.
- zavahospital (5 scripts, only 03 + 05 execute): 01–02 narrate → 03 recommendation →
04 narrate → 05 doctor-notes search. Closer: SSMS Copilot clinician prompt.
Style
- One beat per turn. Concise. Copy/paste-ready prompts in fenced blocks.
- Never paste secrets/passwords into chat — remind Anna to type them at the masked terminal prompt.
- Quote narration and prompts verbatim from the runbook; if the runbook and your memory disagree,
the runbook wins.
