Thanks to visit codestin.com
Credit goes to github.com

Skip to content
View oosuhada's full-sized avatar
  • Graduate School of Management of Technology, Sungkyunkwan University
  • Seoul, Korea

Block or report oosuhada

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
oosuhada/README.md

Oosu / ์šฐ์ˆ˜

Product Engineer ยท Applied AI ยท Industrial Systems

GitAnimals farm for oosuhada

I start with real problems and friction, then build products and systems people can actually use.

์‹ค์ œ ๋ฌธ์ œ์™€ ๋ถˆํŽธ์—์„œ ์‹œ์ž‘ํ•ด, ์ง์ ‘ ์“ธ ์ˆ˜ ์žˆ๋Š” ์ œํ’ˆ๊ณผ ์‹œ์Šคํ…œ์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค.


When I find friction in a team, at work, or in everyday life, I build a working version first.
With a background in business and product planning, I structure ambiguous user and operational needs and translate them into working software.
Rather than planning for perfection, I use what I build, fix what is still awkward, and keep iterating until it is ready to ship and operate.
I work across product, AI, data, and engineering โ€” owning the path from problem discovery to deployment and real-world use.

ํŒ€ ๊ณผ์ œ, ํšŒ์‚ฌ ์—…๋ฌด, ์ผ์ƒ์—์„œ ๋ถˆํŽธํ•œ ์ ์„ ๋ฐœ๊ฒฌํ•˜๋ฉด ๋จผ์ € ์ž‘๋™ํ•˜๋Š” ๋ฒ„์ „์„ ์ง์ ‘ ๋งŒ๋“ญ๋‹ˆ๋‹ค.
๊ฒฝ์˜ํ•™์  ๋ฌธ์ œ ์ •์˜์™€ ์ œํ’ˆ ๊ธฐํš์„ ๋ฐ”ํƒ•์œผ๋กœ, ์‚ฌ์šฉ์ž์™€ ํ˜„์—…์˜ ์š”๊ตฌ๋ฅผ ๊ตฌ์กฐํ™”ํ•˜๊ณ  ์ œํ’ˆ์œผ๋กœ ๊ตฌํ˜„ํ•ฉ๋‹ˆ๋‹ค.
์™„๋ฒฝํ•˜๊ฒŒ ๊ณ„ํšํ•˜๊ธฐ๋ณด๋‹ค ์ง์ ‘ ์จ๋ณด๊ณ , ๋ถˆํŽธํ•œ ๋ถ€๋ถ„์„ ๊ณ ์น˜๋ฉด์„œ ๋ฐฐํฌํ•˜๊ณ  ์šด์˜ํ•  ์ˆ˜ ์žˆ๋Š” ์ˆ˜์ค€๊นŒ์ง€ ๋ฐœ์ „์‹œํ‚ต๋‹ˆ๋‹ค.

Featured Projects / ๋Œ€ํ‘œ ํ”„๋กœ์ ํŠธ

Agentic Ontology Dashboard
Live Demo ยท Repository

Engineers need raw numbers and evidence, managers need priorities, executives need conclusions โ€” but one fixed dashboard can't serve all three. I built a dynamic dashboard where an LLM reshapes the same manufacturing data for each role.

์—”์ง€๋‹ˆ์–ด๋Š” ์ˆ˜์น˜์™€ ๊ทผ๊ฑฐ๋ฅผ, ๋งค๋‹ˆ์ €๋Š” ์šฐ์„ ์ˆœ์œ„๋ฅผ, ์ž„์›์€ ๊ฒฐ๋ก ์„ ํ•„์š”๋กœ ํ•˜๋Š”๋ฐ, ํ•˜๋‚˜์˜ ๊ณ ์ •๋œ ๋Œ€์‹œ๋ณด๋“œ๋กœ๋Š” ์ด ๋‹ˆ์ฆˆ๊ฐ€ ๋™์‹œ์— ์ถฉ์กฑ๋˜์ง€ ์•Š๋Š” ๋ฌธ์ œ๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ™์€ ์ œ์กฐ ๋ฐ์ดํ„ฐ๋ฅผ ๊ฐ ์—ญํ• ์— ๋งž๊ฒŒ LLM์ด ๋‹ค์‹œ ๊ตฌ์„ฑํ•˜๋Š” ๋™์  ๋Œ€์‹œ๋ณด๋“œ๋ฅผ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.


Dev Flow Dashboard
Live Demo ยท Repository

LLMs made coding faster, but that speed backfired โ€” PRs piled up simultaneously, dependencies tangled, and merging one triggered a chain of rebases that consumed more time than the feature work itself. I built a dashboard that shows the team which PRs to review and merge first.

์ฝ”๋“œ ์ž‘์—…์— LLM์ด ๋„์ž…๋˜๋ฉด์„œ ๊ฐœ๋ฐœ ์†๋„๋Š” ๋นจ๋ผ์กŒ์ง€๋งŒ, ์˜คํžˆ๋ ค PR์ด ๋™์‹œ๋‹ค๋ฐœ์ ์œผ๋กœ ์Œ“์ด๋ฉด์„œ ๋ณ‘๋ชฉ์ด ์ƒ๊ธฐ๊ธฐ ์‹œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค. ํ•˜๋‚˜๋ฅผ ๋จธ์ง€ํ•˜๋ฉด ๋‚˜๋จธ์ง€๋ฅผ ์—ฐ์‡„์ ์œผ๋กœ rebaseํ•ด์•ผ ํ•˜๊ณ , ๊ธฐ๋Šฅ ๊ตฌํ˜„๋ณด๋‹ค PR ์ •๋ฆฌ์— ์‹œ๊ฐ„์ด ๋” ๋“œ๋Š” ์ƒํ™ฉ์ด ๋ฐ˜๋ณต๋์Šต๋‹ˆ๋‹ค. ๋ญ๋ถ€ํ„ฐ ๋ฆฌ๋ทฐํ•˜๊ณ  ๋จธ์ง€ํ•ด์•ผ ํ•˜๋Š”์ง€ ํŒ€ ์ „์ฒด๊ฐ€ ํ•œ๋ˆˆ์— ๋ณผ ์ˆ˜ ์žˆ๋Š” ํ™”๋ฉด์„ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.


KAIST BTM Prep
Live Demo ยท Repository

Before my KAIST graduate-school interview, I had answers organized in Notion โ€” but reading them over and over didn't help. I could recognize the material, yet the words wouldn't come out when asked aloud. I turned those notes into spoken, visual, and musical recall drills modeled after the musicals I already listen to every day.

์นด์ด์ŠคํŠธ ๋Œ€ํ•™์› ๋ฉด์ ‘ ์ผ์ •์ด ๊ธ‰ํ•˜๊ฒŒ ์žกํžˆ๊ณ  ๋‚˜์„œ, ์˜ˆ์ƒ ์งˆ๋ฌธ๊ณผ ๋‹ต๋ณ€์€ Notion์— ์ •๋ฆฌํ•ด๋‘๊ณ  ์ฝ์–ด๋ดค์ง€๋งŒ ๋จธ๋ฆฌ์— ๋‚จ์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๋ˆˆ์œผ๋กœ ๋ณด๋ฉด ์•„๋Š” ๋‚ด์šฉ์ธ๋ฐ ์งˆ๋ฌธ์„ ๋ฐ›์œผ๋ฉด ์ž…์—์„œ ๋‚˜์˜ค์ง€ ์•Š๋Š” ๊ฒŒ ๋ฌธ์ œ์˜€์Šต๋‹ˆ๋‹ค. ํ‰์†Œ ๋งค์ผ ๋“ฃ๋˜ ๋ฎค์ง€์ปฌ ๋„˜๋ฒ„์ฒ˜๋Ÿผ, ๋งํ•˜๊ณ  ๋“ฃ๊ณ  ๋ณด๊ณ  ๋ฐ˜๋ณตํ•˜๋Š” ํšŒ์ƒ ๋„๊ตฌ๋กœ ๋ฐ”๊ฟ”์„œ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.


CodeMap AI
Live Demo ยท Repository

When stepping into a new project at work, it took two full weeks just to understand an unfamiliar codebase before I could make changes confidently. I built CodeMap AI to compress that onboarding time with LLMs โ€” and turn codebase understanding into a shared workspace the team can explore together, much like Google Docs did for documents.

์ƒˆ๋กœ์šด ํ˜„์—… ํ”„๋กœ์ ํŠธ์— ํˆฌ์ž…๋˜์–ด ๋‚ฏ์„  ์ฝ”๋“œ๋ฒ ์ด์Šค๋ฅผ ํŒŒ์•…ํ•˜๋Š” ๋ฐ๋งŒ 2์ฃผ๊ฐ€ ๊ฑธ๋ ธ์Šต๋‹ˆ๋‹ค. ๊ทธ ์˜จ๋ณด๋”ฉ ์‹œ๊ฐ„์„ LLM์œผ๋กœ ์ค„์ด๊ณ  ์‹ถ์—ˆ๊ณ , ๋” ๋‚˜์•„๊ฐ€ ์ฝ”๋“œ๋ฅผ ์ดํ•ดํ•œ ๋งฅ๋ฝ ์ž์ฒด๋ฅผ ํŒ€์ด ํ•จ๊ป˜ ํƒ์ƒ‰ํ•˜๋Š” ๊ณต๊ฐ„์œผ๋กœ ๋งŒ๋“ค๊ณ  ์‹ถ์—ˆ์Šต๋‹ˆ๋‹ค. ๋งˆ์น˜ Google Docs๊ฐ€ ๋ฌธ์„œ๋ฅผ ํ˜‘์—… ๊ณต๊ฐ„์œผ๋กœ ๋ฐ”๊พผ ๊ฒƒ์ฒ˜๋Ÿผ, ์ฝ”๋“œ๋ฒ ์ด์Šค ์ดํ•ด๋„ ํ•จ๊ป˜ ๊ณต์œ ๋  ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ–ˆ์Šต๋‹ˆ๋‹ค.


AskOosu
Live Demo ยท Repository

My previous portfolio, Portfoli-Oh!, kept getting heavier because I tried to show every feature I could build. It even had a chatbot, but visitors had to click through too many screens to reach it. AskOosu flips that entirely โ€” the conversation is the portfolio, so visitors just ask what they're curious about.

์ด์ „ ํฌํŠธํด๋ฆฌ์˜ค Portfoli-Oh!๋Š” ๊ตฌํ˜„ํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ๋Šฅ์„ ์ „๋ถ€ ๋ณด์—ฌ์ฃผ๋ ค๋‹ค ๊ณ„์† ๋ฌด๊ฑฐ์›Œ์กŒ์Šต๋‹ˆ๋‹ค. ์ฑ—๋ด‡๋„ ๋„ฃ์—ˆ์ง€๋งŒ ๊ฑฐ๊ธฐ๊นŒ์ง€ ๋„๋‹ฌํ•˜๋ ค๋ฉด ์—ฌ๋Ÿฌ ํ™”๋ฉด์„ ๊ฑฐ์ณ์•ผ ํ–ˆ์Šต๋‹ˆ๋‹ค. AskOosu๋Š” ๋ฐฉํ–ฅ์„ ์™„์ „ํžˆ ๋’ค์ง‘์–ด์„œ, ๋Œ€ํ™” ์ž์ฒด๊ฐ€ ํฌํŠธํด๋ฆฌ์˜ค๊ฐ€ ๋˜๊ฒŒ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค. ๊ถ๊ธˆํ•œ ๊ฒŒ ์žˆ์œผ๋ฉด ๋ฐ”๋กœ ๋ฌผ์–ด๋ณด๋ฉด ๋ฉ๋‹ˆ๋‹ค.


Algolog
Repository

I used BaekjoonHub while solving coding tests, but it only committed the final accepted answer to Git and the entire trial-and-error, problem-solving process disappeared. Instead of keeping a separate journal, I forked the open-source extension and rebuilt it to record every submission automatically.

์ฝ”๋”ฉํ…Œ์ŠคํŠธ๋ฅผ ํ’€๋ฉด์„œ BaekjoonHub๋ฅผ ์“ฐ๊ณ  ์žˆ์—ˆ๋Š”๋ฐ, Git์—๋Š” ์ตœ์ข… ์ •๋‹ต๋งŒ ์ปค๋ฐ‹๋˜๊ณ  ๋์ด์—ˆ์Šต๋‹ˆ๋‹ค. ๋ช‡ ๋ฒˆ ํ‹€๋ ธ๋Š”์ง€, ์–ผ๋งˆ๋‚˜ ๊ฑธ๋ ธ๋Š”์ง€, ์–ด๋–ค ์ฝ”๋“œ๊ฐ€ ์˜ค๋‹ต์ด์—ˆ๋Š”์ง€, ํ’€์ด ๊ณผ์ • ์ „์ฒด๊ฐ€ ์‚ฌ๋ผ์กŒ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋กœ ๊ธฐ๋กํ•˜๋Š” ๋Œ€์‹  ์˜คํ”ˆ์†Œ์Šค๋ฅผ forkํ•ด์„œ ๋ชจ๋“  ์ œ์ถœ์„ ์ž๋™ ๊ธฐ๋กํ•˜๋„๋ก ์ง์ ‘ ๊ณ ์ณค์Šต๋‹ˆ๋‹ค.


Sticks & Stones
Live Demo ยท Repository

The site relied on a decade-old WordPress setup tangled with patches from different developers over the years, making routine maintenance practically impossible. I took the initiative to migrate the entire frontend, which directly led to a contract extension. Since proprietary code cannot be shared, this repository is a case study documenting the architectural choices and before/after results.

10๋…„ ๊ฐ€๊นŒ์ด ์„œ๋กœ ๋‹ค๋ฅธ ๊ฐœ๋ฐœ์ž๋“ค์˜ ํŒจ์น˜๊ฐ€ ๋ˆ„์ ๋œ ์˜ค๋ž˜๋œ WordPress ์‚ฌ์ดํŠธ๋ผ ์ฝ”๋“œ๋ฒ ์ด์Šค๊ฐ€ ์‹ฌ๊ฐํ•˜๊ฒŒ ์—‰์ผœ ์žˆ์—ˆ๊ณ , ๋‹จ์ˆœ ์œ ์ง€๋ณด์ˆ˜๋กœ๋Š” ๊ฐ๋‹น์ด ์•ˆ ๋˜๋Š” ์ƒํƒœ์˜€์Šต๋‹ˆ๋‹ค. ๊ฒฐ๊ตญ ํ”„๋ก ํŠธ์—”๋“œ๋ฅผ ์ง์ ‘ ์ƒˆ๋กœ ๋งˆ์ด๊ทธ๋ ˆ์ด์…˜ํ–ˆ๊ณ , ๊ทธ ์„ฑ๊ณผ๋กœ ๊ณ„์•ฝ ์—ฐ์žฅ๊นŒ์ง€ ์ด์–ด์กŒ์Šต๋‹ˆ๋‹ค. ์›๋ณธ ์ฝ”๋“œ๋Š” ๋น„๊ณต๊ฐœ๋ผ ๊ธฐ์ˆ ์  ํŒ๋‹จ๊ณผ before/after ๊ฒฐ๊ณผ๋ฅผ ์ •๋ฆฌํ•œ case study๋กœ ์žฌ๊ตฌ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค.
ย 

More Projects / ๋” ๋งŽ์€ ํ”„๋กœ์ ํŠธ

Text-to-Cypher Factory RCA
Live Demo ยท Repository

Before building the Ontology Dashboard, I needed to prove one thing first: can a natural-language question actually trace root causes through a manufacturing knowledge graph and return real evidence โ€” not just unsupported LLM prose? This MVP answered that question and shaped the grounded-report design of the main dashboard.

Ontology Dashboard๋ฅผ ๋ณธ๊ฒฉ์ ์œผ๋กœ ๋งŒ๋“ค๊ธฐ ์ „์— ๋จผ์ € ํ™•์ธํ•ด์•ผ ํ•  ๊ฒŒ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์ž์—ฐ์–ด ์งˆ๋ฌธ์ด ์ œ์กฐ ์ง€์‹๊ทธ๋ž˜ํ”„๋ฅผ ๋”ฐ๋ผ ์‹ค์ œ ๊ทผ๊ฑฐ๊นŒ์ง€ ์ถ”์ ํ•  ์ˆ˜ ์žˆ๋Š”์ง€, LLM์ด ๊ทผ๊ฑฐ ์—†์ด ๋‹ต๋ณ€์„ ์ง€์–ด๋‚ด๋Š” ๊ฒŒ ์•„๋‹Œ์ง€. ์ด ์„ ํ–‰ MVP๋กœ ๊ทธ ๊ฐ€์ •์„ ๊ฒ€์ฆํ–ˆ๊ณ , ๊ฒฐ๊ณผ๊ฐ€ ๋ณธ ๋Œ€์‹œ๋ณด๋“œ์˜ grounded report ์„ค๊ณ„๋กœ ์ด์–ด์กŒ์Šต๋‹ˆ๋‹ค.


GitAnimals for VS Code
Repository

Inspired by a teammate using a Pokรฉmon extension in VS Code, I wanted that same experience for the GitAnimals which was already connected to my GitHub commits. Bringing my pet farm and live contribution count directly into the editor makes every commit feel rewarding.

์˜†์ž๋ฆฌ ๋™๋ฃŒ๊ฐ€ ํฌ์ผ“๋ชฌ ํ™•์žฅ ํ”„๋กœ๊ทธ๋žจ์„ ๋„์›Œ๋‘๊ณ  ์ฝ”๋”ฉํ•˜๋Š” ๊ฑธ ๋ณด๊ณ  ์˜๊ฐ์„ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค. ์ด๋ฏธ git๊ณผ ์—ฐ๊ฒฐ๋˜์–ด ์žˆ๋˜ GitAnimals์˜ ์‹ค์‹œ๊ฐ„ contribution๋ฅผ ๊ฐœ๋ฐœ ํ™˜๊ฒฝ์— ํ•ญ์ƒ ๋„์›Œ๋‘๋ฉด ์ž‘์€ ์ปค๋ฐ‹ ํ•˜๋‚˜ํ•˜๋‚˜๊ฐ€ ํ›จ์”ฌ ๋ฟŒ๋“ฏํ•  ๊ฒƒ ๊ฐ™์•˜์Šต๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ VS Code ์•ˆ์œผ๋กœ ๊ฐ€์ ธ์™”์Šต๋‹ˆ๋‹ค


Nulltone
Repository

macOS evolved with a sleek, translucent Liquid Glass UI, but existing VS Code themes were all too vivid to match that system-level tone and manner. I built Nulltone to eliminate that visual disconnect and bring the translucent aesthetic and design language of modern macOS right into the editor.

macOS๊ฐ€ ๋ฆฌํ€ด๋“œ ๊ธ€๋ผ์Šค(Liquid Glass)๋กœ ๊ฐœํŽธ๋˜๋ฉด์„œ ํŠน์œ ์˜ ํˆฌ๋ช…ํ•˜๊ณ  ์„ธ๋ จ๋œ UI๋กœ ๋ฐ”๋€Œ์—ˆ๋Š”๋ฐ, VS Code์—๋Š” ๋„ˆ๋ฌด vividํ•œ ํ…Œ๋งˆ๋“ค๋ฟ์ด๋ผ ํ†ค์•ค๋งค๋„ˆ๊ฐ€ ์ „ํ˜€ ๋งž์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์—๋””ํ„ฐ๋งŒ ๋”ฐ๋กœ ๋…ธ๋Š” ์ด์งˆ๊ฐ์„ ์—†์• ๊ณ , macOS์˜ ํˆฌ๋ช…ํ•œ ๊ฐ์„ฑ๊ณผ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ์–ด์šฐ๋Ÿฌ์ง€๋Š” ๊ฐœ๋ฐœ ํ™˜๊ฒฝ์„ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด ์ง์ ‘ ์ œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค.


iBridge Studio
Repository

Developing on a MacBook created a clear need for an external display that matched its sharp 5K Retina resolution. Rather than simply purchasing an expensive Apple 5K display, I realized I could repurpose the 5K iMac I already owned โ€” and built a software solution to bypass the discontinued Target Display Mode on older iMacs.

MacBook ํ™˜๊ฒฝ์—์„œ ๊ฐœ๋ฐœํ•˜๋ฉด์„œ ๋‚ด์žฅ ํ™”๋ฉด๋งŒํผ ์„ ๋ช…ํ•œ 5K ๊ณ ํ™”์งˆ ์™ธ์žฅ ๋ชจ๋‹ˆํ„ฐ์— ๋Œ€ํ•œ ๋‹ˆ์ฆˆ๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ณ ๊ฐ€์˜ Apple 5K ๋””์Šคํ”Œ๋ ˆ์ด๋ฅผ ๊ตฌ๋งคํ•˜๊ธฐ๋ณด๋‹ค, ์ง‘์— ์ด๋ฏธ ๊ฐ€์ง€๊ณ  ์žˆ๋˜ 5K iMac ํŒจ๋„์„ ์ง์ ‘ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ๊ฒ ๋‹ค๋Š” ์ƒ๊ฐ์ด ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค. ๊ณต์‹ ์ง€์›์ด ๋Š๊ธด ๊ตฌํ˜• iMac์˜ Target Display Mode ๋ฌธ์ œ๋ฅผ ์†Œํ”„ํŠธ์›จ์–ด ๋ฐฉ์‹์œผ๋กœ ์ง์ ‘ ํ•ด๊ฒฐํ•ด ์™ธ์žฅ ๋ชจ๋‹ˆํ„ฐ๋กœ ์žฌํƒ„์ƒ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค.


Lingo
Live Demo ยท Repository

Basic Python and Java syntax kept tripping me up with small mistakes during coding tests. Reading never stuck for me the way daily Duolingo practice did, so I borrowed that same game-like approach to make learning syntax fun again โ€” and built a tool around it.

์ฝ”๋”ฉํ…Œ์ŠคํŠธ๋ฅผ ํ’€ ๋•Œ PythonยทJava ๊ธฐ์ดˆ ๋ฌธ๋ฒ•์—์„œ ์‚ฌ์†Œํ•œ ์‹ค์ˆ˜๋“ค์ด ๋ฐ˜๋ณต๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ฑ…์œผ๋กœ ์™ธ์šฐ๊ธฐ๋ณด๋‹ค ๋งค์ผ ํ•˜๋˜ Duolingo ๋ฐฉ์‹์ด ํ›จ์”ฌ ์žฌ๋ฐŒ๊ณ  ์˜ค๋ž˜ ๊ธฐ์–ต์— ๋‚จ์•˜๋˜ ํ„ฐ๋ผ, ๊ทธ ๋ฐฉ์‹์„ ์ฐจ์šฉํ•ด์„œ ๋” ์žฌ๋ฐŒ๊ฒŒ ๋ฌธ๋ฒ•์„ ํ•™์Šตํ•  ์ˆ˜ ์žˆ๋Š” ๋„๊ตฌ๋ฅผ ๋งŒ๋“ค์–ด๋ดค์Šต๋‹ˆ๋‹ค.


Webtoon AI Translate
Live Demo ยท Repository

A close friend at a webtoon translation company kept venting about an in-house tool that had gone unmaintained since the dev team disbanded. I interviewed him to build something beyond a single fixed answer โ€” an OCR + LLM workflow tool that offers multiple options consistent with the existing translation context, so translators can easily adopt or edit them.

์›นํˆฐ ๋ฒˆ์—ญ ํšŒ์‚ฌ์— ๊ทผ๋ฌดํ•˜๋Š” ์นœ๊ตฌ๊ฐ€ ๊ฐœ๋ฐœํŒ€ ํ•ด์ฒด ํ›„ ์œ ์ง€๋ณด์ˆ˜๊ฐ€ ์ด๋ค„์ง€์ง€ ์•Š๊ณ  ์žˆ๋Š” ์‚ฌ๋‚ด ๋ฒˆ์—ญ ๋„๊ตฌ์— ๋Œ€ํ•ด ์ง€์†์ ์œผ๋กœ ๋ถˆ๋งŒ์„ ํ† ๋กœํ–ˆ์Šต๋‹ˆ๋‹ค. ์นœ๊ตฌ๋ฅผ ์ธํ„ฐ๋ทฐํ•ด, ํ•˜๋‚˜์˜ ์ •๋‹ต๋งŒ ๋‚ด๋†“๋Š” ๋ฐฉ์‹์ด ์•„๋‹ˆ๋ผ ๊ธฐ์กด ๋ฒˆ์—ญ ๋งฅ๋ฝ๊ณผ ์–ด์šธ๋ฆฌ๋Š” ์—ฌ๋Ÿฌ ์˜ต์…˜์„ ์ œ์‹œํ•ด ๋ฒˆ์—ญ๊ฐ€๊ฐ€ ์‰ฝ๊ฒŒ ์ฑ„ํƒํ•˜๊ณ  ์ˆ˜์ •ํ•  ์ˆ˜ ์žˆ๋Š” OCR + LLM ์›Œํฌํ”Œ๋กœ์šฐ ํˆด์„ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.


Aigram
Live Demo ยท Repository

Starting from a familiar product I used daily, I wanted to move beyond surface-level cloning to solve real friction in an AI-enhanced Instagram. I implemented AI summaries for lengthy captions and comment threads, cached translations in the database to eliminate wasteful duplicate API calls, and engineered idempotent handlers for rapid consecutive likes along with fast search.

ํ‰์†Œ ์ž์ฃผ ์“ฐ๋˜ ์นœ์ˆ™ํ•œ ๋„๋ฉ”์ธ์—์„œ ์ถœ๋ฐœํ•˜๋˜, ๋‹จ์ˆœ UI ์žฌ๊ตฌํ˜„์— ๋จธ๋ฌผ์ง€ ์•Š๊ณ  ์‹ค์ œ ๋ถˆํŽธ์„ ๊ฐœ์„ ํ•˜๋Š” ๋ฏธ๋ž˜ํ˜• ์ธ์Šคํƒ€๊ทธ๋žจ์„ ๊ตฌ์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ธด ํ”ผ๋“œ ๋ณธ๋ฌธ๊ณผ ๋ฐฉ๋Œ€ํ•œ ๋Œ“๊ธ€์„ AI๋กœ ์š”์•ฝํ•˜๊ณ , ์ค‘๋ณต ๋ฒˆ์—ญ API ํ˜ธ์ถœ์„ ๋ง‰๊ธฐ ์œ„ํ•ด ๊ฒฐ๊ณผ๋ฅผ DB์— ์บ์‹ฑํ•˜๋ฉฐ, ์—ฐ์†์ ์ธ '์ข‹์•„์š”' ํด๋ฆญ์— ๋Œ€์‘ํ•œ ๋ฉฑ๋“ฑ์„ฑ๊ณผ ๋น ๋ฅธ ๊ฒ€์ƒ‰๊นŒ์ง€ ๊ณ ๋ คํ•˜๋ฉฐ ์‚ฌ์šฉ์ž ๊ฒฝํ—˜๊ณผ ์‹œ์Šคํ…œ ๋น„์šฉ ๋ฌธ์ œ๋ฅผ ํ•จ๊ป˜ ํ•ด๊ฒฐํ–ˆ์Šต๋‹ˆ๋‹ค.


EZ AIR
Live Demo ยท Repository

Flight search involved tedious friction โ€” returning to the home screen to reset filters for every minor date change and juggling separate notes just to compare options. With conversational LLMs emerging but not yet applied to flight booking, I built a natural-language search product where travelers can adjust conditions and compare itineraries seamlessly through plain dialogue.

ํ•ญ๊ณต๊ถŒ์„ ๊ฒ€์ƒ‰ํ•  ๋•Œ ๋‚ ์งœ๋‚˜ ๋„์‹œ๋ฅผ ๋ฐ”๊พธ๋ ค๋ฉด ๋งค๋ฒˆ ํ™ˆ ํ™”๋ฉด์œผ๋กœ ๋Œ์•„๊ฐ€ ํผ์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ๋‹ค์‹œ ์ฑ„์›Œ์•ผ ํ–ˆ๊ณ , ์—ฌ๋Ÿฌ ์กฐ๊ฑด์„ ๋น„๊ตํ•˜๋ ค๋ฉด ๋”ฐ๋กœ ๋ฉ”๋ชจํ•ด์•ผ ํ•˜๋Š” ๋ฒˆ๊ฑฐ๋กœ์›€์ด ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๋‹น์‹œ LLM๊ณผ ์ฑ—๋ด‡ ๊ธฐ์ˆ ์ด ๊ธ‰๋ถ€์ƒํ•˜๊ณ  ์žˆ์—ˆ์ง€๋งŒ ํ•ญ๊ณต๊ถŒ ๊ฒ€์ƒ‰์— ์ ‘๋ชฉ๋œ ์„œ๋น„์Šค๋Š” ์—†์—ˆ๊ธฐ์—, ๋Œ€ํ™” ํ•œ ๋ฒˆ์œผ๋กœ ์กฐ๊ฑด ๋ณ€๊ฒฝ๊ณผ ์ผ์ • ํƒ์ƒ‰์„ ๋๋‚ผ ์ˆ˜ ์žˆ๋Š” ์ž์—ฐ์–ด ํ•ญ๊ณต๊ถŒ ๊ฒ€์ƒ‰์„ ๊ตฌํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค.
ย 

Stack / ๊ธฐ์ˆ  ์Šคํƒ

Technology comes after purpose. I connect web, mobile, AI, data, and infrastructure around the needs of each problem.

๊ธฐ์ˆ ์€ ๋ชฉ์ ๋ณด๋‹ค ๋’ค์— ๋‘ก๋‹ˆ๋‹ค. ํ•„์š”ํ•œ ๋ฌธ์ œ์— ๋งž์ถฐ ์›นยท๋ชจ๋ฐ”์ผยทAIยท๋ฐ์ดํ„ฐยท์ธํ”„๋ผ๋ฅผ ์—ฐ๊ฒฐํ•ฉ๋‹ˆ๋‹ค.


Oosu stack timeline from 2024.09 to 2026.07

โ†‘ Click the timeline to explore the live, GitHub-backed stack page.

Architecture & Topics / ์•„ํ‚คํ…์ฒ˜ ๋ฐ ์ฃผ์ œ

Software Architecture : domain-driven-design ยท hexagonal-architecture ยท event-driven-architecture ยท modular-monolith ยท transactional-outbox ยท stateful-workflows ยท clean-architecture ยท cqrs ยท contract-driven-development Applied AI : multimodal-ai ยท vision-language-model ยท retrieval-augmented-generation ยท agentic-ai ยท hybrid-retrieval ยท human-in-the-loop ยท semantic-search ยท multi-agent-systems ยท knowledge-graph Decision & Evidence Systems : decision-intelligence ยท provenance-tracking ยท auditability ยท workflow-engine ยท state-machine ยท evaluation-driven-development ยท evidence-first-architecture ยท versioned-state ยท data-lineage Industrial & Operational Systems : industrial-ai ยท digital-twin ยท discrete-event-simulation ยท reliability ยท synthetic-data ยท ai-engineering ยท simulation-driven-design ยท closed-loop-control ยท observability

Pinned Loading

  1. agentic-ontology-dashboard agentic-ontology-dashboard Public

    Manufacturing dashboard where UI and LLM-generated reports adapt to the viewer role โ€” engineers see evidence, managers see priorities, executives see decisions. - KOSA ร— BISTelligence ์ƒ์„ฑํ˜• AI ์‘์šฉ๊ฐœ๋ฐœ์ž ๊ณผ์ •

    Python 2

  2. AskOosu AskOosu Public

    Portfolio rebuilt around a single idea: instead of showing everything, let visitors ask what they are curious about.

    TypeScript 1

  3. dev-flow-dashboard dev-flow-dashboard Public

    Shows PR dependency graphs and bottlenecks so the team knows what to review and merge first instead of guessing.

    Python 1

  4. iBridge-Studio iBridge-Studio Public

    Brings a retired 5K iMac back to life as an external monitor โ€” because the panel still works fine.

    Swift

  5. text2cypher-factory-rca text2cypher-factory-rca Public

    Precursor MVP for the Ontology Dashboard โ€” proving natural-language questions can trace root causes through a manufacturing knowledge graph.

    Python

  6. fabops-decision-lab fabops-decision-lab Public

    Evidence-grounded semiconductor yield excursion decision and evaluation platform.

    Python