Building an AI Tutor That Actually Knows Your Students

Every learning platform has a chatbot. Ours knows your name.

When we started building the AI component of Druid Learning Hub, we had a choice. We could bolt on a generic chatbot — the kind that gives you the same answer whether you are a 14-year-old in Osogbo or a 35-year-old data analyst in Abuja. Every other platform does this. It is easy. It is fast. It is completely useless.

We chose the harder path.

Context is everything

Our AI companion does not just answer questions. It knows who you are. When a student named Amara at St Charles Grammar School asks for help with a Python for loop, the AI already knows she is in JSS 2, completed Lessons 1 through 4, struggled with list comprehensions, and has been stuck for 4 minutes.

So instead of a textbook definition, it says: "Remember when you used range(1, 10) in Lesson 3? This is similar, but instead of counting numbers, you are going through a list. Think of it like calling the register — you go through each name one by one."

How it works technically

Every AI call includes a rich system prompt built from profile data, learning history, current context, and struggle detection. This context is fetched via a dedicated API endpoint and injected into every interaction.

Proactive, not reactive

The AI companion sits quietly in the corner of every lesson. If a student has been idle for 3 minutes or the terminal shows repeated errors, the AI button starts pulsing. A small message appears: "Need help with this?"

Students who would normally give up in silence get a gentle push at exactly the right moment.

BYOK for free users

Free users add their own Google Gemini API key. All AI calls happen directly from their browser to Google. We never see the key. For logged-in students, the platform's own key handles everything.

— Ibukun Omonijo, Data Druid Tech