1. The Problem (The “Why”)
While preparing for the “Sport for All Instructor” certification in South Korea, I faced a significant personal pain point: uncertainty. The oral exam had ambiguous grading criteria, causing immense stress and making it difficult to know if my preparation was effective. Existing study materials only provided sample answers, not a way to gauge performance.
2. The Solution (The “What”)
To solve this problem, I decided to build my own tool: a mobile-friendly web application that helps users memorize exam questions and, crucially, uses AI to provide an estimated score for their answers. This feature was designed to directly address the core problem of uncertainty and provide users with the confidence they need to succeed.
3. My Role & The Process (The “How”)
I was the sole product manager, designer, and full-stack engineer for this entire project, from initial concept to deployment.
•
Phase 1 (2024 - MVP for My Own Exam Success): To prepare for my own certification exam, I quickly built a CSR-based prototype (React with Vite, Panda CSS). For the data backend, I used a pragmatic, no-cost approach: populating data in Google Sheets, exporting it as JSON using Google Apps Script, and hosting the file on GitHub Gist. The core feature was an AI-powered scoring system developed without a direct API, but through a manual data curation process. I applied prompt engineering principles, creating a detailed set of instructions that commanded AI models (GPT-4o, Claude 3 Opus) to act as strict examiners, and then systematically collected this AI-generated output to build the ‘answer key’ dataset myself. The accuracy of my custom prompts was validated when the AI’s prediction for my exam score was an exact match to my official result of 84/100 points.
•
Phase 2 (Public Release): To help the next cohort of candidates, I re-architected the app for public use with a modern, scalable stack (Next.js, Supabase). Continuing the successful methodology from Phase 1, I updated and refined the original AI prompts to align with the revised 2025 exam criteria. I then used this process to generate a new, comprehensive dataset of scores and feedback, which was imported into the Supabase (PostgreSQL) database to be displayed in the app.
•
Challenges & Iteration: I navigated real-world technical challenges like OAuth implementation and handling in-app browser limitations in Phase 2. This project reinforced valuable product lessons, such as the need for rapid iteration and prioritizing shipping and an MVP over waiting for perfection.