Inspiration
Our education is transforming to online learning and pioneering the use of AI in teaching. This make me to create Scott-e, a virtual tutor designed to help students learn the concepts of motor learning. Scott-e relied on cloud-based LLM APIs embedded within our learning modules, but I knew there was an alternative way to provide personal, real-time guidance.
What it does
To address the challenge of teaching practical skills remotely, I created an interactive dartboard game. Students can practice their throws from anywhere, with their accuracy data captured directly in the browser to help them identify common errors in motor learning.
Using the built-in prompt API, Scott-e lite, on-device nano model, acts as a personal tutor. It analyzes the student's performance data and provides instant, personalized feedback, explaining motor learning concepts in the context of their own actions. It's the one-on-one tutorial, scaled to every student.
How we built it
This project was brought to life through rapid prototyping. With the clear API reference for the built-in AI, I focused on building the core interactive experience first, the dartboard game. Once the game was working, I integrated the prompt API to connect the student's throw data to the on-device AI model, effectively creating the nano size of the tutor, Scott-e lite.
Challenges we ran into
The biggest challenge was in prompt design and ensuring the stability and accuracy of the AI's output. AI doesn't understand the definition of terms and calculation. To solve this, I had to engineer prompts that included:
- Clear definitions of key concepts like "Constant Error" and "Variable Error."
- The raw calculations from the student's dart throws as context.
By providing the AI with this domain-specific knowledge and real-time data, I could guide it to act as a subject matter expert, not just a language model.
Accomplishments that we're proud of
It runs entirely locally! The fact that the AI can output complex calculations for common errors, explain the concepts behind them, and generate accurate, relevant feedback without running cost!
What we learned
I learned that the future of AI in education isn't just about having the biggest models in the cloud; it's about bringing intelligence directly to the learner. On-device AI, accessed through a simple browser API, is a paradigm shift. It solves the critical issues of cost, privacy, and scalability, making it possible for any educator to build and deploy their own specialized AI tutors.
What's next for Scott-e Lite AI Tutor
The next crucial step is to build trust and enable adoption within an institutional setting. I plan to develop a feature that allows university administrators to review, approve, and manage the prompts and AI behaviors used in courses. This governance layer is essential for ensuring pedagogical quality and deploying Scott-e safely and effectively across the entire university.
Log in or sign up for Devpost to join the conversation.