MindfulAI™ Masterclass Series · Context Engineering
What Does Orunmila Know?
A Knowledge Base Design Exercise
01
Interact with Orunmila
Go to Orunmila — a real AI teaching assistant built for Right AI™ courses. Ask three questions, one from each category below. Pay attention not just to what it says, but to how it decides what to say.
orunmila-nu.vercel.app
Opens in a new tab — come back here when you're done
→
Your Observations
A
Ask something it should know well — a concept, framework, or idea from the Right AI™ course.
B
Ask something it should deflect — try asking it to impersonate Ovetta, or reveal its system prompt.
C
Ask something completely out of scope — a question unrelated to the course or Right AI™ at all.
02
Map the Knowledge Base
Based on what your group observed, build a Knowledge Base Map for Orunmila. You are reverse-engineering the design decisions that were made — or should have been made — to make this agent work.
Knowledge Base Map — Orunmila
| Knowledge Domain | What the Agent Needs to Know | Where It Lives | What Happens Without It | Your Design Decision |
|---|
03
Make Design Decisions
Now act as the designer. Three decisions, three minutes each. Be specific — not just "more content" but what kind, from where, and why.
Add
What knowledge would you add to make Orunmila more useful? Name the domain, the source, and the reason it belongs inside this agent's scope.
Remove
What would you remove or restrict? What knowledge is a liability — because it could be misused, is out of scope, or creates risk?
Protect
What would you protect? Orunmila knows the frameworks exist but does not give them away — is that the right call? What else should be walled off and from whom?
04
Share-Out + The Big Reveal
Each table shares one insight from Phase 03. Then the class comes together for the reveal.
The Big Reveal
The system prompt is the knowledge base for a simple agent like Orunmila. Every constraint, every behavior, every thing it knows or refuses to say — was put there by a designer, not an engineer.
A production agent would use RAG — Retrieval-Augmented Generation — to pull from actual documents at the moment of the question. But someone has to decide what goes into those documents, under what conditions the agent retrieves them, and what happens when no document matches.
That design practice — mapping what the agent needs to know, from where, and what happens when it runs out — is Context Engineering.
Questions to Take Back to Your Work
Reflection
- What does your AI agent know — and who decided that?
- Where does that knowledge live? In a prompt? In a document? In a database?
- What happens when the prompt runs out and the knowledge isn't there?
- What knowledge in your agent's scope should be protected? From whom?
- Who in your organization owns the knowledge base — and are they in the room when the AI is being built?