December flew by like a lightning bolt, and now there are only ten days left of the year. 😓 Personally, the biggest win this year was getting used to using AI as a sort of 'external brain.' At the same time, it was a period of constant awe watching how fast LLMs are evolving. I'm originally a non-dev 'Mundori' (liberal arts major) who used to envy developers who could manifest code out of thin air, never dreaming I’d actually touch it myself. Then, around March this year, I heard that 'Gemini 2.5 Pro is great at coding' and 'it lets even non-devs build programs.' That’s how I started, by making a simple text converter.
▲ An image EXIF remover I made in the early days. Even though it was a simple program, seeing real, working code pop out based on my instructions made me go nuts. 👀 Once I realized 'Wait, this actually works,' I figured since I can generate AI images too, I could combine the two to create something. That’s how my eventful 'Vibe Coding' journey began. 😅
1. Live On! Wakaba The first thing I decided to make was... a raising simulation game (!), quite fearless of me. 😅 I shared a post about it here once. I put a lot of thought and hard work into it, but it was a bit of a bummer that there was a barrier at the download stage since I tried to distribute it as a ZIP file for local use. Maybe I can revive it as a web app or mobile format someday.
2. Arcana Code v1.0 Next, I made a Tarot card web app using LLM. The concept was solid and prototyping was fast, but the BYOK (Bring Your Own Key) method—requiring user API keys—held it back. Since it was a web deployment, I figured there'd be trust issues with users entering their API keys... so it ended up being an awkward local app. I'm currently rebuilding this as a Chrome extension. I want to add multi-language support and go global. I'm hoping to drop an MVP on the Web Store early next year. haha
▲ Screen from Arcana Code 2.0 currently in development
3. Ether Tuner The third thing is a conceptual app combining brown noise + radionics in a static website format. 😅 Finished it on May 17th. I was obsessed with brown noise back then and wanted to listen to it without having to open a YouTube video. I made it so the visualizer and the brown noise change subtly based on keywords, but Gemini 2.5 struggled so hard with the visualizer. If I remade it with recent models, I think I could do it much better.
4. Prompt Viewer This is a Chrome extension that lets you easily see prompts for AI images (NAI, Stable Diffusion) via a right-click menu. According to the stats, it has a whopping 1,600 users... it's the most successful app I've made in terms of user base. Though it didn't make a dime since there's no monetization. 😅 The last update was June 4th, so I must have made it before early summer.
5. Archive 03 I've introduced this on Cliang before as well. It's a blog that serializes AI-generated finance content in Korean/English, with a concept where three anime-girl personas have some back-and-forth 'tiki-taka.'
For the content generation, I've automated HTML formatting and English localization for free using the Gemini CLI. However, since AI... sometimes does great research but then hallucinates and talks total nonsense, a final human review and polish is a must. Also, since I picked finance as the topic (a topic Google is very strict about), it’s not easy to show up in search results. lolol
6. Webplified I worked really hard on this one too. It's a tool that converts PNG/JPG files to WebP. The last update was July 5th. It’s completely free with no ads, and looking back, I’m not even sure why I spent so much time polishing it. 🙄 Anyway, it’s live, and I still use it often when I want to shrink image sizes. haha
7. Haruna On Screen The big one: 'Haruna On Screen,' a screenmate for Windows. I shared this on Cliang too. I started making it around July/August and just pushed the 1.0.4 update yesterday. In yesterday's update, I added a 'school uniform' outfit and included the Gemini 3 Flash model in the options. Fixed some minor bugs too. I originally wanted to release it on Steam, but it got rejected due to external API usage issues, so I ended up putting it on itch.io (an indie software platform) for $2.99. Since I was able to implement almost exactly what I envisioned, it's one of the outputs I'm quite proud of. 😊 Yesterday, I even got feedback from a user saying they're using it to study Japanese/Chinese, which made me feel super accomplished.
8. Prognos 'Prognos,' a daily US stock chart predictor. A 'homemade' AI I trained at home with a single GPU. Yeah... writing this out, I realize I really made all sorts of stuff. This wasn't actually because I wanted to predict stock prices, but rather a learning project that started with the question: 'Can you really do machine learning at home with a 3060 GPU??' With Gemini's help, I applied architectures like LSTM and GRU and churned out over a dozen models that actually predict daily candles. Thanks to that, I think I've gained some understanding of machine learning. Of course, the brain size of the resulting models was at the level of a flatworm (like a leech or something)... lolol
9. Line by Line: A Timeline of My Thoughts This is another Chrome extension. It's a minimalist app where you record your daily life and thoughts line by line, and the AI reads them to give you feedback. Since it replaces the Chrome 'New Tab' page, the big advantage is that you can just jot things down mindlessly every time you open a tab. Out of all the apps I've made, 'Line by Line' is the one I use most frequently. The 'kick' (cool feature) is that it implements cross-device sync using Google Drive as a backend without needing a server. ✌ (Zero maintenance cost!) The feedback persona in the screenshot is a bit unique, but you can choose other 'normal' (?) characters too. 😅 The latest version 1.2.5 is currently up on the Chrome Web Store. It even has a feature that analyzes your strengths and weaknesses and gives advice based on long-term records. I also added Gemini 3 Flash to the options recently. haha
10. Wrapping up. Man, I didn't think this would get so long. To be honest, besides what I've written here, there are about six or seven other projects sleeping on my hard drive that I stopped midway through planning or implementation. 😅😅 (In the midst of all this, I even released an EP of AI music. lol) I have two or three ideas waiting to be built right now, and thanks to model improvements and tools like Anti-gravity, I expect to churn out various works even faster than before. Anyway, it feels like a year where I worked really hard creating things despite having a day job. Thinking about what's become possible thanks to AI over the past year... all I can say is that it's surprisingly amazing. Next year, I hope to create something that actually generates meaningful revenue through vibe coding. I'll end this here, wishing everyone a great end of the year and a healthy, happy New Year. 😊 Thank you!
6. 웹플리파이드
이것도 굉장히 열심히 만들었던 거네요. PNG/JPG 파일을 WebP 파일로 변환해주는 도구입니다. 마지막 업데이트 날짜는 7월 5일입니다. 완전 무료에 광고도 없는데 왜 그렇게 열심히 깎았는지 지금 생각해 보면 다소 의문이군요. 🙄 아무튼 공개는 되어 있고, 저도 이미지 용량 줄이고 싶을 때 종종 애용하는 물건이네요. ㅎㅎ
7. 하루나 온 스크린
대망의 그 물건, 윈도용 스크린메이트 '하루나 온 스크린'. 역시 클리앙 소개글로 한번 공유를 했었습니다. 대략 7~8월경에 만들기 시작했고, 바로 어제 1.0.4 업데이트를 진행했지요. 어제 업뎃에선 '교복' 의상을 추가하고 제미니 3 플래시 모델을 모델 선택지에 넣었습니다. 자잘한 버그도 잡았고요. 본래 스팀에 출시하려 했지만 외부 API 사용 이슈로 리젝되었고, 결국 itch.io라는 인디 소프트웨어 플랫폼에 2.99달러로 올렸습니다. 제가 생각한 것을 거의 그대로 구현할 수 있었기 때문에, 나름 자부심을 가지고 있는 산출물 중 하나입니다. 😊 어제는 일본어/중국어 학습용으로 사용하고 계시다는 사용자분의 피드백도 들을 수 있어서 무척 뿌듯했네요.
8. 프로그노스
미국 주식 일봉 차트 예측기 '프로그노스'. 집에서 글카 한 장으로 훈련시킨 수제(?) 홈메이드(?) AI입니다. 네... 쓰다 보니까 제가 진짜 별 걸 다 만들었네요. 이건 사실 진짜 주가를 예측하고 싶었다기보다는 3060 GPU로 진짜 집에서 머신러닝이 가능해??? 라는 의문으로 시작했던 학습용 플젝이었음니다. 제미니의 도움으로 LSTM, GRU 등 아키텍처를 적용해 실제로 일봉을 예측하는 모델을 십수 가지나 뽑아 봤네요. 덕분에 기계학습에 대한 이해는 어느 정도 생겼던 것 같습니다. 물론 만들어진 모델의 뇌 크기는 기껏해야 편형동물(거머리 같은 거) 수준이긴 했지만요... ㅋㅋㅋㅋ
9. 한줄한줄: 내 생각의 타임라인
이건 또 다른 크롬 익스텐션으로, 평소에 한 줄씩 내 일상과 생각을 기록해 두면 AI가 그걸 읽고 피드백을 주는 미니멀한 앱입니다. 크롬 새 탭을 대체하기 때문에, 새탭 열 때마다 아무 생각 없이 그때그때 마구 기록을 써제낄 수 있다는 큰 장점이 있습니다. 실제로 제가 만든 앱 중에 직접 사용하는 빈도가 가장 높은 것이 바로 이 '한줄한줄'인데요. 서버 없이, 구글 드라이브를 백엔드처럼 사용해서 기기 간 기록 동기화를 구현한 부분이 킥입니다. ✌ (유지비 0원!) 캡처짤에 피드백 페르소나가 좀 독특한데요. 쟤 말고 좀 다른 정상적인(?) 캐릭터도 선택 가능합니다. 😅 현재는 크롬 웹스토어에 1.2.5 최신 버전이 올라가 있습니다. 긴 기간의 기록을 바탕으로 아래와 같이 사용자의 장단점을 분석하고 조언해 주는 기능도 들어가 있지요. 최근 제미니 3 플래시 모델도 선택지에 넣어뒀습니다. ㅎㅎ
10. 마무리. 아니, 이렇게 길어질 줄은 생각 못했네요. 사실 지금 쓴 거 말고도 기획이나 구현 중간에 멈춘 플젝들이 제 하드에 예닐곱 가지는 잠들어 있습니다. 😅😅 (그 와중에 AI 음악으로 EP 음원을 출시하기도 했죠. ㅋㅋㅋ) 지금도 두세 가지 아이디어들이 구현을 기다리고 있는데, 모델 발전과 안티그래비티 덕분에 전보다 훨씬 빠르게 다양한 작업물을 만들 수 있을 것으로 기대됩니다. 아무튼 따로 본업이 있는 상태였음에도 정말 열심히 이것저것 만들어댄 한 해였던 것 같습니다. 지난 1년간 AI 덕분에 뭐가 가능해졌는지 생각해 보면... 새삼스럽지만 놀랍다고밖에 할 말이 없네요. 내년에는 바이브 코딩으로 의미 있는 수익을 내는 무언가도 꼭 만들어 보고 싶다는 바람입니다. 그럼 여기까지 읽어주신 여러분께서도 한 해 잘 마무리하시고, 건강하고 행복하게 새해 맞이하시길 바라며 마칩니다. 😊 감사합니다!
"The community is in awe of this 'Mundori' powerhouse who's out-coding the pros, while an IT veteran gives a nod of respect to the new era of AI-powered creativity."
#SupportiveContinue Browsing
![[NSFL] I've started hating AI since the New Year](https://kcomm-assets.b-cdn.net/thumbnails/thumb_0daa16c5df60cf7c.webp)
