My Experience at Kent Hack Enough

Participating in Kent Hack Enough was one of the most exhilarating and educational experiences of my college career so far. For those unfamiliar, Kent Hack Enough is Kent State University's annual hackathon where students come together to build innovative projects within a 24-hour timeframe.

The Beginning: Forming Our Team

Our journey began with team formation. I partnered with Connor Love, who shared my passion for educational technology. We both believed that technology could revolutionize how people learn, making education more accessible and engaging for everyone.

Our small but effective team consisted of:

• Me (TJ) - AI generation and STEVE development
• Connor Love - Frontend development and demo work

The Idea: Edu-Synapse

After brainstorming several concepts, we settled on creating Edu-Synapse, an adaptive learning platform that personalizes educational content based on individual learning styles and progress. The core features included:

• Personalized learning paths that adapt to user performance
• Interactive content with immediate feedback
• Spaced repetition algorithms to optimize memory retention
• Collaborative learning spaces for peer-to-peer interaction

STEVE: Our AI Learning Assistant

The core of our project was STEVE (System for Teaching, Evaluating, and Visualizing Education), an AI-powered learning assistant designed and implemented during the 24-hour hackathon. STEVE leverages a transformer-based architecture fine-tuned for educational content generation and personalized learning experiences. While a general LLM, STEVE has been specifically trained to align with learning science principles, resulting in superior performance on pedagogical tasks. On internal benchmarks, STEVE achieves parity with OpenAI's gpt-3.5-turbo on metrics like knowledge retention and concept application. Furthermore, it rivals Anthropic's Claude 3.7 Sonnet on benchmarks evaluating the clarity and coherence of educational explanations, and knowledge integration across multiple sources. STEVE functions both as an interactive chatbot and as a standalone content generator for constructing comprehensive learning paths.

STEVE's core functionalities include:

• Personalized Learning Paths: STEVE creates custom learning journeys by analyzing your performance and interests. It tracks your progress, uses collaborative filtering to recommend resources, and builds paths based on your knowledge gaps and preferred learning styles. It uses a dynamic "Knowledge Graph" to understand the subject matter and tailor the learning experience.
• Adaptive Content Explanation: Complex topics are broken down into smaller, manageable concepts. Each concept is explained step-by-step, with the length and complexity of the explanations adjusted to your skill level. This adjustment is based on your past interactions and a statistical model, fine-tuning the language and level of detail used.
• Automatic Content Analysis and Integration: STEVE can process various document formats (PDFs, DOCX, etc.) and automatically extract key concepts and relationships using techniques like Named Entity Recognition and Keyphrase Extraction. This extracted information is then added to the Knowledge Graph, expanding STEVE's understanding of the subject.
• Real-time Information Retrieval: STEVE can search the web for up-to-date information using a combination of keyword and semantic search strategies. Retrieved information is then ranked and filtered based on source credibility, topic similarity, and your preferences. Relevant excerpts are incorporated into the learning content to provide context.
• Code Generation and Explanation: STEVE can generate Python code to solve problems, analyze data, and illustrate concepts. The code is executed in a secure environment, and the results are interpreted and explained. This is particularly useful for learning math, statistics, and computer science, utilizing a specialized "Code Interpreter" component.
• Learning Support Material Generation: STEVE can generate study notes, outlines, and summaries to help users accelerate learning. It can vary the type of summary generated from bullet point notes to structured outlines depending on the request from the user.

A key differentiating factor for STEVE is its adaptive resource allocation strategy. STEVE dynamically allocates "thinking steps" (internal computation cycles) based on query complexity. For simple queries, it provides fast, concise responses. For more complex tasks, it allocates additional resources to perform more in-depth reasoning, resulting in more detailed and comprehensive explanations. This "thinking steps" approach is inspired by research on sample-efficient reasoning capabilities in large language models, specifically those utilizing chain of thought prompting and hierarchical reasoning techniques. This is implemented via conditional prompting and re-ranking techniques.

The Challenge: 24 Hours of Coding

Once we had our concept, the real challenge began. We had just 24 hours to bring our idea to life. We quickly established a workflow, dividing responsibilities based on our strengths while ensuring continuous communication.

I focused on designing and implementing STEVE's AI capabilities, including the core reasoning engine, web searching functionality, and Python code generation and execution features. Meanwhile, Connor worked on creating an intuitive frontend interface and preparing the demo that would showcase our platform's capabilities.

The final hours were a race against time. We focused on implementing our core features and ensuring everything worked together seamlessly. We also prepared our presentation, knowing that how we communicated our idea would be just as important as the technical implementation.

The Presentation: Showcasing Our Work

When presentation time arrived, we were exhausted but excited. We had created a functional prototype that demonstrated the potential of our idea. Our presentation focused on three key aspects:

1. The problem: Traditional education's one-size-fits-all approach
2. Our solution: Personalized, adaptive learning with STEVE
3. Demo: A live demonstration of our platform, including generation of content specified by the judges.
4. Questioning period: A one-on-one discussion with each judge to evaluate our technical knowledge, and to answer further questions about the platform.

For the demo, Connor showcased the frontend while I demonstrated STEVE's capabilities. We showed how STEVE could:

• Process a complex math problem, break it down into manageable steps, and adapt its explanation based on the user's skill level
• Generate Python code to solve a programming challenge, execute it, and interpret the results
• Search the web for relevant information to enhance its explanations with up-to-date resources
• Adapt the learning path based on user performance, adjusting content difficulty in real-time

Despite some technical hiccups during the demo, such as STEVE not knowing what pieces are called in chess, the judges seemed very impressed with our concept and implementation. The questions they asked showed genuine interest in how our solution could be applied in real educational settings.

The Result: Third Place!

When the results were announced, we were thrilled to learn that we had secured third place! The recognition validated our hard work and belief in our idea. Beyond the prize, the real reward was the experience itself--the skills we developed, the connections we made, and the confidence we gained.

Lessons Learned

Participating in Kent Hack Enough taught me several valuable lessons:

• The power of a small, focused team: With just two people who worked well together, we accomplished more than we thought possible in 24 hours.
• The importance of scope management: We had to be realistic about what we could accomplish in our limited time.
• The value of resilience: Pushing through fatigue and setbacks was essential.
• The benefit of preparation: Having some familiarity with the tools we planned to use saved precious time.

The Future of Edu-Synapse

The hackathon might be over, but our journey with Edu-Synapse continues. We're currently refining our prototype and exploring opportunities to develop it further.

You can join the Edu-Synapse platform today at https://edu-synapse.com/signup

We believe in the potential of our platform to make a real difference in education. By combining adaptive learning paths with STEVE's personalized assistance, we're creating an educational experience that truly adapts to each learner's needs.

If you're interested in learning more about Edu-Synapse or following our progress, visit https://edu-synapse.com.

Final Thoughts

If you're a student considering participating in a hackathon, my advice is simple: do it! The experience is invaluable, regardless of the outcome. You'll learn new skills, make connections, and push yourself in ways you never thought possible.

And who knows? You might just create something amazing in the process.