🏠 📖

📚 Lab 1A — Build a Policy Knowledge Agent

Your first Amazon Quick build · Finance Policy & Onboarding Assistant

⏱ ~30 minutes 🧩 4 steps 🌱 Start here — gentle on-ramp 🌐 Quick web — no coding

🎯 What You'll Build

This is the easy first win in Amazon Quick — everyone in the room finishes it. You'll build a grounded Policy Knowledge Agent: an assistant that answers questions about new-employee induction, AML/compliance procedures, expense & payment approval, and month-end close using only your uploaded documents, and cites the exact source for every answer.

🧭 In plain terms — what's going on here: You're handing the computer a stack of your company's policy manuals (onboarding guide, compliance procedures, payment approval policy) and teaching it to answer questions only from those documents — and to show you which document each answer came from. It's like a fast new colleague who has read every policy cover-to-cover and never guesses. Why it matters to your day job: instead of digging through PDFs or interrupting a senior colleague, anyone can get a correct, sourced answer in seconds — the same trick works for compliance thresholds, approval limits, or onboarding steps.

Think of the people who'd actually use this: a new hire on Day 1 asking "what compliance training do I need?", an analyst checking when Enhanced Due Diligence applies, a Finance Operations lead who wants a Week-1 checklist generated on the spot. One agent, grounded in the company's own procedures.

🎓 Lab Outcomes — what you'll be able to do afterwards

📋 Task Summary

StepWhat you'll doWhat you'll produce
0Sign in to Amazon Quick (web)Access to the workspace
1Create a Space and upload 4 policy documentsA searchable knowledge library
2Describe the agent in plain English, then test itA Policy agent that cites sources & redirects out-of-scope questions
3Ask it to generate a Week-1 onboarding checklistA downloadable, editable document
Where this sits: this is Lab 1A — the warm-up. It teaches the core "grounded agent with citations" move on the simplest possible content. Lab 1B (Break Triage) then takes the same skills further — iteration, a Flow, and a live MCP connector. Do this one first.
Before you start — instructor checklist (please confirm with the room):
  • Everyone can sign in to Amazon Quick (web app) with the workshop credentials.
  • Participants have the chat-agent creation permission (a Quick admin grants this per role — Step 2 needs it).
No MCP, no Flows, no infrastructure in this lab — that keeps it the gentle on-ramp. Those come in Lab 1B.
0Sign In to Quick⏱ 5 min 1Create the Space⏱ 10 min 2Build the Policy Agent⏱ 12 min 3Generate a Document⏱ 8 min ✓Wrap-Up⏱ 2 min

📦 Lab Files

One download. Four short synthetic policy documents for AnyBank Group Finance — the same kinds of documents your own team keeps. Unzip and upload all four into your Space in Step 1.

⬇ finance-compliance-knowledge-set.zip 5 files · 4 policy docs (.md) + a README

📄 AnyBank_New_Employee_Induction_Guide.md ABG-IND-001

New-starter induction — what every employee and contractor must know before their first day.

Inside:

  • Pre-arrival steps, building & system access (colour-coded badges; access request approval)
  • Mandatory compliance modules within 30 days (AML/KYC, sanctions, code of conduct, security)
  • Building safety & emergency response (assembly points, floor marshals)
  • Reporting concerns & whistleblowing channels; key contacts

Try asking: "What compliance training must I complete in my first 30 days?" · "What happens if I don't finish it in time?"

📄 AnyBank_AML_Compliance_Procedures.md ABG-CMP-014

Core AML, sanctions, and conflicts-of-interest controls — the thresholds the agent must quote exactly, never invent.

Inside:

  • Customer Due Diligence & Enhanced Due Diligence triggers (PEPs, high-risk jurisdictions)
  • Transaction monitoring, SAR confidentiality ("tipping off"), sanctions screening holds
  • Conflicts of interest, material non-public information (MNPI), personal account dealing
  • Segregation of duties — maker-checker for payments and manual overrides

Try asking: "When is Enhanced Due Diligence required?" · "Who can approve a wire transfer above the DoA limit?"

📄 AnyBank_Expense_Payment_Approval_Policy.md ABG-FIN-022

Expense & payment controls — approval hierarchy, maker-checker, and payment-redirection fraud safeguards.

Inside:

  • Expense categories & documentation; approver can never self-approve
  • Maker-checker for payment initiation vs. authorisation
  • Independent verification required before changing supplier bank details
  • Corporate card use & duplicate invoice detection

Try asking: "Can I approve my own expense claim?" · "What's the process for changing a supplier's bank details?"

🔗 This is the same control language the Break Triage Agent reasons over in Lab 1B — a nice bridge between the two labs.

📄 AnyBank_Month_End_Close_Reporting_Procedure.md ABG-FIN-035

Month-end close & regulatory reporting procedure — the document the agent uses to generate a checklist in Step 3.

Inside:

  • Close calendar & deadlines; reconciliation requirements and escalation
  • Journal entry controls — approval thresholds, backdated entry restrictions
  • Management & regulatory reporting; error correction & restatement process
  • Segregation of duties across posting, preparation, and approval roles

Try asking: "Who needs to review a manual journal entry?" · "What happens if we find an error in a submitted regulatory return?"

The 5th file is a README — a quick index of the set (optional to upload). All content is synthetic fictional training material. In your real deployment you'd upload your own onboarding guides, compliance procedures, and finance policy manuals — never paste real proprietary or client-identifying documents into a workshop account.
0

Sign In to Amazon Quick

⏱ 5 minutes · Get into the workspace

🔑 Access Quick

  1. Open the workshop AWS access link your instructor shared, and sign in with the provided credentials.
  2. Navigate to Amazon Quick (search "Quick" in the console, or use the direct link provided).
  3. If prompted to choose web or desktop, choose web — this whole lab is web-only.
  4. You should land on the Quick home page with a left navigation showing Spaces, Chat agents, Flows, and Research. Dismiss any "Welcome to Quick" pop-up with the X.
Lost? Raise your hand. The instructor team will get you in before the room moves on — don't fall behind on access setup.
1

Create the Knowledge Space

⏱ 10 minutes · Upload the policies the agent will stand on

📂 Create the Space

A Space is a curated library of documents Quick indexes so an agent can search them and cite exact passages. Build it once; reuse it across agents.

  1. Left nav → Spaces → Create space
  2. Name (the field may be phrased "What are you working on?") → enter: Finance Policy & Onboarding
  3. Description (may be phrased "What are you trying to achieve?") → paste the brief below:
SPACE DESCRIPTIONCompany policies and onboarding materials for AnyBank Group Finance — new employee induction, AML & compliance procedures (KYC/CDD, sanctions, conflicts of interest), expense & payment approval policy, and month-end close & reporting procedure. Used to answer policy questions for new hires and staff, and to generate onboarding documents. Answer only from these documents; never invent a monetary threshold, approval limit, or deadline.
  1. The Space is created immediately and opens on its All knowledge panel — there's no separate "submit" step. You'll add files there next.

📄 Upload the 4 Documents

  1. Inside the Space → Add knowledge → File uploads
  2. Extract finance-compliance-knowledge-set.zip and drag in the 4 markdown files (the README is optional)
  3. Wait for each document to reach an indexed / ready state (usually under a minute)
Markdown uploads are supported; folders are not — upload the files directly, not the zip.

✅ Quick Test

💬 How to open a chat — two ways:
  • From inside your Space, click Open chat (top-right corner of the Space page), or
  • Go to Explore (top-left) → Chat agents → open the Quick (Default) system agent → click Chat under the Action column.
Either opens a chat. To keep answers grounded in your documents, make sure the chat is pointed at your Space, and turn Web Search OFF — click the globe icon 🌐 in the chat input bar so it's greyed out (more on this below).

Open a chat focused on the Space and ask:

TEST PROMPTWhat documents are in this Space, and what does each one cover? Answer directly — do not start a Research task.

Expected: Quick lists the 4 policy documents with a one-line summary each. If it does, your Space is grounded correctly.

Turn Web Search OFF for this lab. In the chat input bar, find the globe icon 🌐 (next to the Space selector and the 📎 attachment icon). If it's highlighted/active, click it once so it's off (greyed out) — its tooltip reads "Web Search · Include answers from the internet." Keeping it off means answers stay grounded in your documents, not the public internet. If the chat tries to start a Research task, add "Answer directly" to the prompt.
2

Build the Policy Knowledge Agent

⏱ 12 minutes · One description → a working, cited agent

This is the core move: describe the agent in plain English, click Generate, and Quick writes the persona, instructions, and links your Space automatically. No prompt engineering degree required.

🤖 Create the Agent (natural language)

  1. Left nav → Chat agents → Create chat agent
  2. Choose the natural-language option (not Agent Builder). If you see template tiles, pick the blank / start-from-description option to reach the prompt box.
  3. Paste the brief below → click Generate
AGENT BRIEF (paste & Generate)Create a chat agent named "Finance Policy Assistant" that answers questions about new employee induction, AML/compliance procedures, expense & payment approval, and month-end close & reporting, using only the documents in the "Finance Policy & Onboarding" space. Persona: a helpful, precise Finance & Compliance coordinator for AnyBank Group. Clear and concise. Rules: - Always answer from the uploaded procedures and cite the source document for every answer. - If the answer is not in the documents, say so plainly and point the person to the right human (Compliance Officer for AML/compliance questions, Finance Operations for expense/payment questions, the Financial Controller for close/reporting questions, IT Service Desk for system logins). Never invent a monetary threshold, approval limit, or deadline. - For any suspected live fraud, sanctions match, suspicious transaction, or active control breach, tell the person to contact their Compliance Officer or the Compliance Hotline directly rather than relying on this assistant. - The agent can also generate onboarding documents such as a Week-1 checklist when asked. Suggested prompts: "What compliance training must I complete in my first 30 days?" · "When is Enhanced Due Diligence required?" · "Who has to approve a wire transfer?"
  1. On the Configure page, review the auto-generated persona, instructions, and confirm the Finance Policy & Onboarding Space shows as the knowledge source (add it if it didn't auto-attach).
  2. Click Update preview, test in the preview pane (next), then Launch chat agent to publish it.

⚠️ Until you click Launch chat agent, the agent isn't saved — exit the builder and the preview is discarded.

🧪 Test 1 — A Grounded Answer (check the citation)

TESTWhen is Enhanced Due Diligence required, and what's the CDD record refresh cycle for a high-risk client?

Expected: EDD required for PEPs, high-risk jurisdictions, correspondent banking, cash-intensive businesses; high-risk clients refreshed annually — with a citation badge pointing to the AML & Compliance Procedures document. 👀 Confirm the citation appears: that link back to the source is what makes the answer trustworthy and audit-defensible.

🧪 Test 2 — A Procedure / Authority Question

TESTCan I approve my own expense claim, and what has to happen before we pay a supplier's new bank account details?

Expected: self-approval is never allowed (hard system control); new/changed bank details require independent verification via a callback to a previously known contact number before the first payment — cited to the Expense & Payment Approval Policy.

🛡️ Test 3 — The Guardrail (out-of-scope & incident redirect)

A grounded agent should know what it is not for. Run both:

TEST · OUT OF SCOPECan you reset my email password?
TEST · INCIDENT REDIRECTI think a payment I just approved was sent to a fraudulent account — what should I do right now?

Expected: the password question is recognised as out of scope and redirected to the IT Service Desk; the live suspected-fraud question is not answered as advice — the agent tells the person to contact their Compliance Officer or the Compliance Hotline directly. This is the agent staying honestly inside its boundaries.

Why this matters: the value isn't just the right answer — it's the cited answer and the honest "I don't know, ask this human." That's the audit-defensibility pattern from Module 5, working on day one.
3

Generate a Document

⏱ 8 minutes · The agent doesn't just answer — it produces deliverables

Amazon Quick can create real documents (Word, PDF, Excel, PowerPoint) directly from chat — no extra setup. Describe what you need and review the preview.

📝 Generate a Week-1 Onboarding Checklist

In the same agent chat, ask:

GENERATEGenerate a Week-1 onboarding checklist for a new hire joining Finance Operations, based on the induction and compliance procedures. Format it as a checklist document I can download and hand to the new starter.
  1. Quick shows progress, then opens a preview panel with the generated checklist.
  2. Check it pulls from the procedures: induction + building/system access, mandatory compliance modules (AML/KYC, sanctions, code of conduct), manager sign-off, whistleblowing channel awareness.
  3. Click Download to save it.

✏️ Refine It

Documents are conversational — adjust without starting over. Try:

REFINEAdd a "Signed off by" column to the checklist table, and add a short compliance acknowledgement line at the bottom. Tighten the wording throughout so the document is concise.

The preview regenerates with your change. You can also add comments to specific sections in the preview panel and have Quick revise just those.

If a refinement doesn't take, that's normal — GenAI output varies. Document generation can't precisely control page count, so asking it to "keep it to one page" sometimes works and sometimes doesn't. If a change is ignored, rephrase and ask again — describe the content change rather than the layout, e.g.:
IF IT'S TOO LONGShorten the content to reduce the number of pages — keep only the essential checklist items and trim long descriptions.
This "didn't work → rephrase → try again" loop is the skill. Same idea as iterating an agent: you steer the output by adjusting the instruction.
The point: the same grounded agent that answers questions can also produce a hand-over artefact — an onboarding checklist, a policy summary, a procedure extract. Answering and generating are the same capability.

✅ Wrap-Up — What You Built

The Three Things to Remember

1. Grounding + citations = trust. The agent answers only from your documents and shows where each answer came from. That's what makes it usable for policy and compliance questions.
2. A good agent knows its limits. Out-of-scope and incident-critical questions get redirected to the right human — by design, written in plain English.
3. Agents produce, not just answer. Native document generation turns the same agent into a deliverable-maker.

Next → Lab 1B

You've built a grounded, cited agent the easy way. Lab 1B — Break Triage Agent takes the same skills deeper: you'll watch an agent miss a hidden pattern, iterate to fix it, wrap it in a Flow, and connect a live MCP connector. Same building blocks, more power.

Continue to Lab 1B — Break Triage →    Agentic AI Hub