⬡ Overview
📋 Brief
🔌 Connectors
🗄 Data Plan
🔒 Privacy
💻 Live IDE
◇ Check
★ Results
0 / 5
Hands-On Lab · Module 10 · BAA 10.0

Secure
Integration Project

You're the architect. Design a real data integration plan for an AI agent that reads customer data from a live source — then defend every decision in front of a privacy auditor. The last stop: a live IDE where your agent actually runs.

What You'll Build
🔌
Connector Plan
Choose and justify the connectors your agent needs — CRM, spreadsheet, or both.
🗄
Data Access Map
Specify exactly which fields the agent reads — and argue why each one is necessary (or not).
🔒
Privacy Audit
Identify and document privacy concerns — then propose mitigations a regulator would accept.
💻
Live IDE
Paste your API key and run your agent for real — see it respond to queries about the integration you designed.
Lab Roadmap
—
Read the Brief
Your client, their use case, and the agent's mission
1
Choose Your Connectors
Select from CRM, Google Sheets, and hybrid options — match to the use case
2
Map the Data Fields
Tag each field as Necessary / Sensitive / Should Not Access
3
Privacy Concern Analysis
Identify the top 4 privacy risks — describe mitigations
4
Live IDE — Run Your Agent
Enter an LLM API key, write your system prompt, and test the agent live
5
Knowledge Check · 7 questions · 70% to pass
The Client Brief

Northstar SaaS Co.

Your client is a B2B SaaS company with 3,200 customers. They want an AI agent that reads their customer data and helps their Customer Success team answer questions like: "Which accounts are at risk of churning?" and "Who hasn't logged in for 30 days?"

Client Situation

Northstar's Customer Success (CS) team of 12 currently pulls weekly reports manually from two systems: their HubSpot CRM (account records, contact info, deal history) and a Google Sheet maintained by the ops team (manual churn scores, NPS survey results, support ticket counts, last-login dates from their product database export).

The CS team wants an agent they can ask natural-language questions to. The agent should read both sources, find relevant accounts, and surface insights without the CS manager having to write a single formula.

CRM (HubSpot)
  • 👤 Contact: name, email, phone, title
  • 🏢 Company: name, industry, size, ARR
  • 💼 Deal: stage, close date, MRR, owner
  • 🗒 Notes: CS rep call notes (free text)
  • 📅 Activity: email opens, meeting history
  • 🔐 Billing: payment method type, status
Google Sheet (Ops Export)
  • 🕐 Last login date (from product DB)
  • 📊 Monthly active users (MAU) count
  • 🎯 NPS score (0–10, last survey)
  • 🎫 Support tickets opened (last 90d)
  • ⚠️ Churn risk score (manual, 1–5)
  • 📝 CS notes (free text, ops additions)
⚠️ What makes this hard: The data contains both operational metrics (fine to read) and personal information (needs justification). Some fields the agent could access have no legitimate purpose for churn analysis — but they're in the same source. Your job is to design the integration precisely, not permissively.
Section 01 · Design Activity

Choose Your Connectors

Select the connectors the agent needs. Then answer the justification question — why this combination, and not a simpler or broader one?

💡 Think like an architect: More connectors = more attack surface. A connector that touches PII but doesn't provide unique signal should be reconsidered. Pick what you need — no more.
Available Connectors
Section 02 · Design Activity

Data Access Map

Every field the agent reads is a liability as well as an asset. Tag each field below. Then submit — the correct answers reveal the principle behind data minimization.

✅ Necessary — agent needs this to do its job
⚠️ Sensitive — useful but requires justification
🚫 Should Not Access — no legitimate agent need
Section 03 · Privacy Analysis

Privacy Concern Analysis

You must identify and document at least one privacy concern in the integration you just designed. Click each concern to assess it and propose a mitigation. A real data integration plan requires all four.

🔒 The rule: Every AI agent that touches PII must have a written privacy analysis before it goes to production. This is not optional under GDPR, CCPA, or most enterprise security policies. This section is that document.
Reviewed: 0 / 4
Section 04 · Live Lab

Live Integration IDE

Your agent design is now ready to run. Enter an API key from an LLM provider, customize your agent's system prompt, and test it with real queries about the integration you built.

🔑 API Key Safety: Your key is stored only in this browser session — it is never transmitted to any server except the LLM provider's official API endpoint directly from your browser. You can use any Anthropic, OpenAI, or compatible key. Clear the field when done.
🔑 LLM Connection
API Key
Not connected — enter your API key above
📄 System Prompt Editor
system_prompt.txt
System Prompt
Data Schema
Privacy Rules
💬 Agent Test Console
CS
Northstar CS Agent
Ready for queries
claude-sonnet-4-20250514
CS
Hello! I'm your Northstar Customer Success agent. I can answer questions about customer health, churn risk, and account activity based on your CRM and ops data. Connect your API key to activate me.
Try These Queries
Knowledge Check

7 Questions

Drawn directly from the integration design and privacy principles in this lab. 70% to pass.

Question 1 of 7Score: 0 / 0
Project Debrief

Project Complete