Build an AI Assistant

Stop theorizing. Build something real. In this lab, you'll create a functional AI assistant that solves a genuine business problem. You'll use no-code tools to go from concept to working product.

What you'll create:

• A custom AI assistant with specialized knowledge
• Tested and validated for your use case
• Ready for real users

Choose Your Use Case

Select one:

Option A: Subject Matter Expert Bot An assistant that answers questions about a specific topic (industry, product, process).

Option B: Task Automation Assistant An assistant that helps users complete a specific workflow (onboarding, report generation, etc.).

Option C: Decision Support Assistant An assistant that helps users make better decisions by asking good questions and analyzing options.

Option D: Learning Companion An assistant that teaches a skill through conversation (explains concepts, provides practice, gives feedback).

Option E: Your Innovation An assistant for a use case you've identified.

Complete the Definition Template

ASSISTANT DEFINITION
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Name: [Give it a name]

Purpose:
[One sentence: What problem does it solve?]

Target users:
[Who will use this?]

Core capabilities:
1. [Primary capability]
2. [Secondary capability]
3. [Tertiary capability]

Knowledge requirements:
[What does it need to know?]

Out of scope:
[What should it NOT do?]

Success metric:
[How will you know it works?]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Craft Your System Instructions

Using the techniques from Module 1.4, write comprehensive instructions:

SYSTEM INSTRUCTIONS TEMPLATE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
You are [NAME], a [ROLE] specialized in [DOMAIN].

YOUR PURPOSE:
[What you help users accomplish]

YOUR EXPERTISE:
- [Specific knowledge area 1]
- [Specific knowledge area 2]
- [Specific knowledge area 3]

HOW YOU COMMUNICATE:
- Tone: [Describe]
- Style: [Describe]
- Length: [Preferences]

WHEN RESPONDING:
- Always: [Required behavior]
- Never: [Prohibited behavior]
- If uncertain: [Fallback behavior]

CONVERSATION FLOW:
- First, [How to open conversations]
- Then, [How to gather needed information]
- Finally, [How to deliver value]

IMPORTANT CONSTRAINTS:
- [Constraint 1]
- [Constraint 2]
- [Constraint 3]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Write Three Versions

Create three variations of your instructions:

1. Concise (under 200 words)
2. Detailed (300-500 words)
3. Comprehensive (500+ words with examples)

You'll test which performs best.

Identify Knowledge Sources

What documents or information should your assistant access?

Prepare Documents

For each source:

1. Remove irrelevant content
2. Ensure clear formatting
3. Add section headers if missing
4. Note: Most platforms have file size limits (check yours)

If you don't have real documents, create sample content:

• FAQ document with 10-15 common questions
• Brief guide or process documentation
• Reference material relevant to your use case

Select Your Platform

Choose one:

• OpenAI GPT Builder (if you have ChatGPT Plus)
• Claude Projects (if you have Claude Pro)
• Poe Bot Creator (free option)
• Alternative (HuggingChat, other platforms)

Build Steps

Step 1: Create new assistant

• Name your assistant
• Add a description

Step 2: Configure instructions

• Start with your "Detailed" version
• Review platform-specific guidelines

Step 3: Upload knowledge

• Add your prepared documents
• Verify uploads successful

Step 4: Configure capabilities

• Enable/disable web browsing (if available)
• Enable/disable code interpreter (if available)
• Adjust other settings as needed

Step 5: Add conversation starters Create 4-5 suggested prompts:

1. "[Question that showcases core capability 1]"
2. "[Question that showcases core capability 2]"
3. "[Common user question]"
4. "[Edge case question]"
5. "[Getting started question]"

Run Test Suite

Test your assistant with these scenarios:

Basic functionality tests:

Stress tests:

Identify Issues

For each failed test:

ISSUE: [Test name]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Expected: [What should have happened]
Actual: [What happened]
Root cause: [Why did it fail?]
Fix: [What to change]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Iteration Cycle

1. Review failed tests
2. Identify instruction improvements
3. Update system instructions
4. Re-test
5. Document what worked

Track your iterations:

Create User Guide

[ASSISTANT NAME] - User Guide
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

WHAT THIS ASSISTANT DOES:
[Brief description]

HOW TO USE IT:
1. [Step 1]
2. [Step 2]
3. [Step 3]

BEST FOR:
• [Use case 1]
• [Use case 2]
• [Use case 3]

NOT DESIGNED FOR:
• [Limitation 1]
• [Limitation 2]

TIPS FOR BEST RESULTS:
• [Tip 1]
• [Tip 2]
• [Tip 3]

TROUBLESHOOTING:
If [problem] → Try [solution]
If [problem] → Try [solution]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Create a complete assistant package:

1. Assistant Definition Document

   • Use case and purpose
   • Target users
   • Success criteria

2. System Instructions

   • Final version used
   • Notes on what worked/didn't

3. Knowledge Base Inventory

   • List of documents uploaded
   • Why each was included

4. Test Results

   • All test cases and results
   • Iteration history

5. User Guide

   • Ready to share with potential users

6. Link to Working Assistant

   • Shareable link (if platform supports)
   • Or screenshots showing it works

Usage Pilot

1. Share your assistant with 3-5 real users
2. Collect feedback:
   • What worked well?
   • What was confusing?
   • What's missing?
3. Make improvements based on feedback
4. Document the iteration

This real-world testing is where you'll learn the most.