EulaIQ

Inspiration

I got into pharmacy at 16, a year early, at one of Nigeria's top universities. I did well in my first year, but I dropped out later because I wanted to build software, not study medicine.

Before EulaIQ, I built two other AI products on my own. The first turned textbooks into podcasts and quizzes. Google released NotebookLM for free and it did what my product did, better and free, so that was the end of that. The second turned lessons into AI-generated animations. It won a statewide hackathon with 700+ teams and got real paying users, but the video quality was too low and educators dropped off over time.

Both failures taught me the same thing. Students in Nigeria preparing for exams like JAMB and WAEC don't just need information, they need something that actually teaches them, and reacts to them when they're confused. A textbook can't do that. A video can't either, no matter how good it looks, because it can't hear you when you're stuck.

What it does

EulaIQ turns notes and textbooks into a live class taught on an AI-powered whiteboard. An AI instructor teaches the material visually, step by step, and you can interrupt at any point, ask a question, answer a question it asks you, and it picks the lesson back up exactly where you left off.

How I built it

The frontend is Next.js, hosted on Vercel. The server runs on Node.js and handles the core orchestration. LLM calls and the agent's reasoning run on Bedrock, this is what decides what to teach, when to ask a question, and how to respond mid-lesson. Speech-to-text runs on AWS STT, so the system can hear the student when they jump in. Text-to-speech runs on ElevenLabs, that's the instructor's voice. The whiteboard itself is animated with GSAP, building the visual explanation step by step in real time.

The core technical breakthrough came from my second failed product. That animation tool used one-shot generation, ask the model for a video, get back whatever it gives you, no way to correct it. That's why the quality was inconsistent and why it eventually failed. For EulaIQ, I moved to an iterative system instead, the AI calls tools step by step, gets feedback from the rendering engine, and adjusts as it goes. That's what makes the whiteboard reliable enough to teach with, and it's the same idea that now powers real-time interruption, since the system is already built to react and adjust one step at a time instead of committing to one long output.

Challenges I ran into

Getting real-time interruption working was the hardest part. The system has to keep teaching, listen for the student at any moment, understand what they asked, and resume the lesson without losing where it was, all without noticeable lag. Cost was another constant pressure, real-time generation and voice are expensive to run well, and I had to think hard about what could be cached, pre-rendered, or simplified without hurting the teaching itself. And I did all of this solo, no technical cofounder, so every one of these problems was mine alone to work through.

What I learned

I learned that good pedagogy isn't just about content quality, it's about responsiveness. My second product had the right idea, animated teaching, but no way to react to a confused student. Fixing that taught me that the interruption model isn't a feature bolted onto a video, it has to be the foundation the whole system is built around from the start.