GeoLearn AI

Inspiration

As undergraduate STEM students in Rwanda, our inspiration came from a frustrating, daily reality. Rwanda’s Vision 2050 relies on building a knowledge-based economy driven by engineers, geologists, and agricultural scientists. Yet, there is a massive "Resource Gap" in our education system. Most secondary schools and universities cannot afford to build million-dollar physical geological laboratories or maintain comprehensive collections of rock and mineral samples.

We found ourselves memorizing textbook definitions of "Granite" and "Coltan" without ever touching them. When students graduate and enter the field, they struggle to physically identify the minerals right at their feet. We realized that if we couldn't bring every student into a physical lab, we needed to put a digital laboratory into the pocket of every student. That is how GeoLearn AI was born.

What it does

GeoLearn AI is a mobile-first, web-based EdTech platform that acts as a "Digital Pocket Laboratory."

It solves the resource gap through three core features:

The AI Scanner: Students use their smartphone camera to scan any rock or mineral. Our custom computer vision model instantly identifies the sample, providing its exact mineral composition and scientific description.

The AI GeoTutor: We integrated a conversational chatbot directly into the results page. Instead of just giving the student a static answer, they can ask the AI follow-up questions like, "How did this rock form?" or "Where is this found in Rwanda?"

The Dynamic Learning Hub: A secure, no-code Admin CMS allows educators to instantly update a localized database of rocks and minerals. The Hub automatically features "Rock of the Month" and "Mineral of the Week" to keep students engaged outside of fieldwork.

How we built it

To maximize accessibility, we avoided heavy native app development and built GeoLearn AI as a responsive, lightweight web application. We focused heavily on Technical Execution and User Experience, ensuring the app felt premium and native even on low-end smartphones.

Frontend Architecture: We utilized HTML5, vanilla JavaScript, and Tailwind CSS to build a clean, glass-morphism UI with interactive tab navigation.

The AI Brain: We trained our own computer vision classification model using Google's Teachable Machine. We then integrated the exported model directly into our frontend using TensorFlow.js.

The Database & State Management: To keep the prototype fast and serverless, we engineered a dynamic JavaScript object database that acts as a Content Management System (CMS). Educators can log in via a secure PIN vault, upload images, add resources, and the app instantly re-renders the DOM to update the Learning Hub and search indexing.

Challenges we ran into

Our biggest technical challenge was balancing AI confidence thresholds with user experience.

Initially, if the TensorFlow model was less than 80% confident about a rock scan (due to poor lighting or a low-quality smartphone camera), the app would fail to return a result. This created a frustrating UX.

We had to fundamentally rethink our logic. We adjusted the probability arrays so that if the model's confidence dropped below 30%, it gracefully degraded into a "System Unknown" state. However, instead of stopping there, we ensured the AI GeoTutor chatbot remained active even on failed scans. This allowed the student to describe the rock manually to the AI, turning a technical failure into a conversational learning moment

Accomplishments that we're proud of

We are incredibly proud that we did not just build a "scanner," we built a complete, scalable EdTech ecosystem. The integration of a functioning Admin portal alongside the AI model proves that GeoLearn AI has genuine Business Viability (B2B institutional licensing). We proved that high-impact African tech solutions don't require million-dollar budgets; they require resourceful founders.

What we learned

We learned that the best educational technology doesn't replace the teacher; it scales them. By building the AI GeoTutor and the dynamic database, we learned how to build software that adapts to localized curricula (like featuring Rwandan Coltan) rather than forcing Western-centric data onto African students.

What's next for GeoLearn AI

In the immediate future, we plan to migrate our local database to a cloud infrastructure like Firebase to enable multi-user accounts and data persistence. Long-term, our goal is to expand the TensorFlow model beyond geology to include agricultural soil analysis, empowering not just students, but local farmers across the continent.