TEACHLOOP HACKATHON SUBMISSION
Inspiration
TeachLoop was inspired by a personal experience during group study sessions.
When I studied alone, I often ended up memorising information without truly understanding it. But when I studied with friends, something changed. We constantly explained concepts to each other, asked questions, corrected mistakes, and forced each other to think more deeply.
At first, I assumed this improvement came from simply studying together. Over time, I realised the real reason: we were learning by teaching.
Explaining a concept exposed gaps in understanding immediately. If I could not explain something clearly, I did not fully understand it. When someone challenged my explanation, I had to rethink it properly instead of relying on memorisation.
This became even more obvious during exam preparation, where questions are often reworded, twisted, or designed to test understanding in unfamiliar ways. Knowing a concept is not enough, you need to understand it under pressure and in different contexts.
That experience led to a simple question: What if every student had someone to teach, anytime?
TeachLoop was built from that idea.
⸻
What it does
TeachLoop is a learning system built around a continuous teach and learn loop where students improve understanding by actively teaching an AI.
Instead of functioning as a traditional tutor that simply provides answers, TeachLoop shifts roles dynamically depending on how the user is learning:
- Teach Mode: the AI explains a concept step-by-step and helps build foundational understanding
- Student Mode: the AI becomes a confused learner, asking questions, misinterpreting explanations, and challenging the user to clarify their thinking
Student Mode is also designed to simulate real exam conditions where confusion and ambiguity are common. The AI intentionally reframes concepts in slightly misleading or twisted ways, similar to how exam questions test understanding in unfamiliar contexts. This forces users to go beyond memorisation and develop flexible understanding that can be applied under pressure.
If explanations are incomplete or incorrect, the AI identifies gaps and guides refinement rather than simply providing the answer.
Users can also switch topics within the same session or separate them into different chats depending on how they prefer to study.
TeachLoop supports learning from real materials. Users can upload documents such as PowerPoints or notes, and the AI uses them as contextual grounding to evaluate explanations and check accuracy.
The result is a continuous learning cycle: Learn → Teach → Challenge → Correct → Repeat until mastery
⸻
How we built it
TeachLoop is a web-based AI prototype designed to demonstrate a structured learning interaction rather than a standard chatbot.
We built a prompt-driven system that controls AI behaviour across two modes: Teach Mode and Student Mode. The AI dynamically switches roles based on user interaction, creating a structured learning loop instead of static question-and-answer responses.
We integrated file upload functionality so users can provide learning materials such as PowerPoints or notes. The AI uses this content as context when evaluating explanations and identifying misconceptions.
The system was built using:
- Meta AI for core implementation
- ChatGPT and GitHub Copilot for system design and development support
- OpenCode for additional coding assistance
- Vercel for deployment
- GitHub for version control
The focus was intentionally on delivering a functional MVP that clearly demonstrates the learning loop rather than a production-ready system.
⸻
Challenges we ran into
The biggest challenge was time. As high school students, we were building TeachLoop under significant time pressure while balancing exams and other commitments, forcing us to carefully prioritise which features to include in our MVP.
Another major challenge was implementing the AI itself. Finding the right APIs, managing API keys, integrating AI models into our application, and ensuring reliable communication between the frontend and backend required extensive experimentation, debugging, and testing.
One of the most technically demanding aspects was engineering TeachLoop’s dual AI system. Unlike traditional AI assistants that immediately provide answers, we developed two distinct AI behaviours: Student Mode, where the AI learns from the user by asking questions, making realistic mistakes, and encouraging deeper thinking, and Teacher Mode, where it explains concepts, identifies misconceptions, and provides constructive feedback. Designing these behaviours while keeping conversations natural and human-like required significant prompt engineering and iteration.
Implementing real-time voice conversations was another major challenge. We had to learn how voice AI systems work, integrate speech technologies, manage audio input and output, and ensure conversations felt responsive and natural. Building a smooth voice experience that complemented both AI modes added another layer of technical complexity.
Finally, bringing the overall concept to life was challenging because TeachLoop explores a new approach to AI-assisted education. Transforming the idea of “learning by teaching an AI” into an intuitive, engaging, and functional product required continuous experimentation, problem-solving, and refinement throughout the hackathon.
⸻
What we learned
This project reinforced that effective learning systems are not defined by how much information they provide, but by how they force users to think.
We also learned how powerful interaction design and prompt architecture are in shaping user behaviour, especially in AI-driven educational tools.
Small structural changes in how an AI responds can completely change the depth of user understanding.
⸻
What’s next for TeachLoop
TeachLoop is currently a functional MVP focused on demonstrating the core learning loop.
Future development includes:
- more refined AI personality switching between modes
- improved evaluation of user explanations
- classroom dashboards for teachers to track understanding
- personalised learning paths based on user performance
- expanded support for structured curriculum content
The long-term vision is to give every student an AI learning partner they actively teach, transforming passive studying into active mastery.
Log in or sign up for Devpost to join the conversation.