Build & Test (Customer PoC)
Test your solution
After generation completes, test the agent:
Use the Generated Sample Questions
Each scenario generates ready-to-use sample questions in the data folder:
The file contains three categories of questions:
| Category | Source | Example |
|---|---|---|
| Structured data questions | Fabric IQ (data) | "How many open claims do we have?" |
| Unstructured data questions | Foundry IQ (docs) | "What's our process for filing a property claim?" |
| Combined questions | Both sources | "Which claims are approaching our SLA deadline based on our process guidelines?" |
Use these questions first
The generated questions are tailored to your scenario's data and documents. Start with these before improvising.
Testing Tips
Start with structured data questions
Show the power of natural language over structured data:
"How many [entities] do we have?"
"What's the total [metric] for [time period]?"
"Show me the top 5 [entities] by [metric]"
Then unstructured data questions
Demonstrate intelligent document retrieval:
"What's our policy on [topic]?"
"How do we handle [process]?"
"What are the requirements for [action]?"
Finish with combined questions
This is the "wow" moment: questions that need both sources:
"Based on our [policy], which [entities] need attention?"
"Are we meeting our [documented SLA] according to the data?"
"Which [items] don't comply with our [policy/guidelines]?"
Prepare your test script
Before customer meetings, prepare 5-7 questions:
| # | Question Type | Example |
|---|---|---|
| 1 | Structured data | "How many open claims do we have?" |
| 2 | Structured data | "What's the total value of claims filed this month?" |
| 3 | Structured data | "Which agents have the most policies?" |
| 4 | Unstructured data | "What's our process for filing a property claim?" |
| 5 | Unstructured data | "What does our standard homeowner policy cover?" |
| 6 | Combined | "Which open claims are approaching our SLA deadline based on our process guidelines?" |
| 7 | Combined | "Do any current claims involve coverage types not in our standard policy?" |
Let customers ask questions
After your prepared questions, let customers ask their own questions. This shows the solution handles real scenarios, not just scripted ones.
Launch the Web Application
The web application is already deployed during the initial azd up deployment. Open the app URL shown in the deployment output in your browser to interact with the agent through a chat interface.
Checkpoint
Ready for Customer PoC
Your custom PoC should:
- Answer structured data questions accurately
- Retrieve relevant unstructured document content
- Combine sources for complex questions
- Use industry-appropriate terminology
Next: Review Deep dive to prepare for technical questions
Quick Reference: Regenerate for another industry use case
# Edit .env with new customer's industry and use case, then:
python scripts/00_build_solution.py --clean
# Or inline:
python scripts/00_build_solution.py --clean \
--industry "New Industry" \
--usecase "New use case description"