Over the past year, I’ve been running a personal experiment: using AI tools to go through the full cycle of validation → productization → commercialization — solo, on weekends.
Tools like n8n, Claude Code, and Replit made this possible for someone like me who can’t write code from scratch.
At work (ShopBack), AI has already been deeply embedded in how we operate — I won an internal AI Innovation Award last year for process automation, and our team now uses Claude Code to accelerate product development. But what surprised me more was what happened when I brought the same mindset into my personal life.
Over the past six months, I built 4+ apps. Each one taught me something different about where the real difficulty lies.
Four Projects. Four Different Kill Switches.
- Baobao (Insurance Policy Organizer): Technically feasible, but I hit a wall fast — regulatory complexity and resource barriers in the insurance industry. My kill criterion was clear: when compliance costs far outweigh the value of MVP validation, it’s not an optimization problem. It’s a “wrong path for now” problem. Q_Q
- PocketFast (Fasting Tracker): I’m genuinely happy with the product — it’s fun and does what it should. But the market is saturated, and reaching competitive feature parity would take more resources than I have right now. Still a project I enjoy. Kill criterion: competitor density × differentiation headroom.
- Iju no Nihongo (NHK News Self-Study): Functionally solid, but copyright and content scaling are a hard ceiling — not something effort can solve. It’s a structural business model constraint, and being honest about that is more valuable than pushing through. Q_Q (Feel free to try it though!)
These “failures” clarified something for me: in the AI era, the hard part isn’t building — it’s finding the right battlefield.
The barrier to execution is lower. The bar for PMF hasn’t moved.
But with clear kill criteria, the cost of being wrong is also much lower. 😛
Why LinguaPass?
LinguaPass is a study tool for Taiwan’s Ministry of Education Certification for Teaching Chinese as a Foreign Language (TOCFL Educator Exam). A few PM-driven reasons I chose this:
- Institutional demand with clear boundaries. This exam is tied directly to career pathways and government subsidies — it’s not “learn Chinese for fun.” It’s a licensing-type need that doesn’t fade with trends.
- Niche but well-defined market. Since its launch in 2006, the exam has reached its 21st edition, with ~35,670 cumulative applicants and 6,631 certified. In 2025 alone, 1,669 people sat the exam. Small numbers, but extremely high intent — exactly the kind of market where study efficiency and structured learning create real value.
- I was the user. Like many candidates Q_Q, I had to study by manually checking PDF answer keys — impossible to practice during commutes, and not enough to build a real test strategy.
- The moat is knowledge depth, not tooling. With 10+ years observing the exam from the teaching side, plus a PhD in linguistics, I can structure the exam logic into a learning system — not just aggregate questions.
Product Design: Starting from Test-Taker Psychology
My UXR background means I default to user psychology, not feature lists. The core of LinguaPass:
- Systematic answer logic. Getting the answer right matters less than understanding why — aligned to exam logic and question patterns. That’s the real value of a study product, and the hardest part to build well.
- Error analysis and tagging. A weakness-tracking system built on linguistics-informed categorization — turning years of exam trends into something teachable, practiceable, and analyzable.
- Reducing friction. Connecting structured knowledge with fragmented study sessions, keeping the “I want to practice” barrier as low as possible on mobile.
Where Things Stand
LinguaPass officially launched on April 25, 2026. Less than three weeks in, with zero marketing, I already have both regular free users and real paying customers.
In PM terms: I’ve hit the paid validation milestone. That doesn’t mean PMF is fully confirmed — but it means the value proposition and business model are pointing in the right direction.
What keeps me more motivated than revenue: users are proactively telling me what to build next.
I started focused on linguistics subjects, but user feedback pushed toward expanding to other exam sections — which means I now have a clear product expansion path, and real prioritization decisions to make.
A Small Closing Thought
The name LinguaPass took a while to land: Lingua, Pass, and a Taiwanese pun on “PhD” (波酥 / bōsoo) — I was way too happy when it clicked 😛
It didn’t come from prompting an AI. It came from a family brainstorm, old-fashioned human taste.
The tools are more accessible than ever. But taste still determines whether what you ship feels like your product. 😛
I’ll keep sharing: pricing strategy, metric design, and how to balance content quality with iteration speed in a niche market.
If you’re building something verifiable with AI, or curious about the UXR × PM × AI intersection — let’s talk.
