How 41 Teams Built Real AI Projects in Just 4 Weeks

AI education programs are everywhere — but how many of them actually produce working results? That was the core question behind the Sogang University AI Learnathon, a 4-week intensive co-hosted by Codepresso and Sogang University from January 19 to February 12, 2026.

The answer: 41 teams submitted functioning projects, and 45 out of 63 participants completed the program — a 71.4% completion rate.

Why Most AI Training Falls Short

Many AI programs focus heavily on content delivery. Participants learn concepts, complete quizzes, and walk away with certificates. But there is often a gap between understanding AI and actually building something with it. Knowledge without application rarely sticks.

Codepresso's learnathon model was designed to close that gap. Instead of separating learning, assessment, and project work into isolated phases, the program weaves them into a single continuous flow. Every lesson feeds into a hands-on task, and every task feeds into a real project.

The Program Structure That Drove Results

The 4-week program combined online self-paced learning with live sessions to accommodate different skill levels. The final day brought participants together for an offline workshop and hackathon-style presentations, where teams demonstrated their completed projects.

A key enabler was bkend, an AI-native Backend-as-a-Service platform by PopupStudio. Participants used bkend to handle database creation, API connections, and authentication — the essential backend components that typically block beginners from shipping a product.

Two Tracks, One Goal

For technical participants, the AI Agent track guided learners from prompt engineering through knowledge-base integration to multi-step agent workflows using LangChain and LangGraph. The goal was clear: move from explaining an idea to demonstrating a working prototype.

For non-technical participants, the Vibe Coding track proved that the right tools can unlock real capability. Using Cursor and bkend, teams with no prior development experience built fully functional web services — complete with databases, APIs, and user authentication. What started as skepticism quickly turned into momentum once ideas became live, clickable products.

What Makes AI Education Stick

The most important metric for any training program is not how much content it covers. It is what participants can do afterward. When learners walk away with a deployed project, a presentation they delivered, and the confidence that comes from finishing what they started — that is education that lasts.

Codepresso's learnathon model is built on a simple principle: design for completion, not just participation. Structure the experience so that every learner has a realistic path to a finished product. The 71.4% completion rate and 41 shipped projects at Sogang University show what becomes possible when program design prioritizes outcomes over content volume.

Interested in running a results-driven AI education program for your university or organization?