How can Sunghyun help?

# How can Sunghyun help?

I want to know how Sunghyun Cho can help with my project. Sunghyun owns cho.sh, a personal site with a digital garden of notes and a set of mini-apps. Everything below about him is public. Read this whole prompt, research him with the access listed in it, read my project at the end, and write the answer described under `<answer>`.

<profile>
Sunghyun wrote this block in the first person. Paths that start with / in it are on https://cho.sh.

# Sunghyun Cho

Who I am and what I do

Hey. I am Sunghyun Cho. I am a Search ML Engineer at TwelveLabs. I studied CS at USC on a Boeing Scholarship and have shipped products at Woowa, Karrot, Grammarly, and Lunit.

## Building search at TwelveLabs as a Search ML Engineer.

Building cho.sh, my own encyclopedia and super-app.

## Dossier

### 1

I consider myself an accelerationist. Every earthly problem traces back to not enough energy[^1], so humanity's first job is to command more of it. So to speak, an energy version of the Malthusian trap[^2] is encroaching. But traps don't call for a solution; they call for a breakthrough[^3], and I think fusion is ours. Lately I've been deep in stellarators. A reactor shaped like the Baekje Gilt-bronze Incense Burner would look so badass.

[^1]: I treat energy as fungible with food, resources, money, and desire

[^2]: Malthusian trap is a pessimistic theory Thomas Malthus laid out in An Essay on the Principle of Population: population grows geometrically while food grows only arithmetically, so shortage, poverty, and collapse keep recurring.

[^3]: Solutions solve a problem; breakthroughs make it moot. The last Malthusian trap ended not with rationing but with Haber-Bosch.

### 2

We often overestimate what we can do in a day and underestimate what we can do in a year. I'm also a victim of this. Once you've built and shipped an app in a single day and watched it hit 230,000 weekly active users[^4], you never quite stop overestimating a day. So lately I keep relearning the value of plans and deadlines. Speed matters not because of time to market but because willpower is the first resource to run out, and you need real metrics before it does.

[^4]: 750,000 visits, 500,000 installs. Then MikeRoweSofted!

### 3

I've been super into health these days. I never quite got why the rich spend millions on health, but now I do. They've bought everything fate offered except the one thing they can't, a body. So to speak, a person is a carbon computer, and consciousness is the higher-order thing running on top of it. Once you split body from mind, you consider yourself a soul trapped in outdated hardware. The two can't actually be fully separated, so all you can do is take care of the machine. Easier said than done. My own [hardware-maintenance report](/mini/health) is embarrassing.

### 4

I'm an aggressive early adopter. I pay premiums for something unfinished because I'm buying not the object but a white paper on the makers' philosophy. So to speak, I buy ideas, not products; an insource of bleeding-edge ideas keeps my own thinking sharp. Same for softwares. For instance, I first tried Notion in 2018[^5], and I've been on Obsidian since 2022.

[^5]: I mostly remember how slow and buggy it was. It took me two years to come back.

### 5

I consider myself a humanist. The economy doesn't come first with people slotting into jobs to serve it. So to speak, work gets defined by human things, time, focus, responsibility, cooperation, confidence, stamina; the economy is what falls out of that. Therefore, [just as cities grew to the radius of a one-hour commute](/0F3111), the economy grows to the radius of what a person can do in a day. Technology widens free will, so tasks that used to be too hard get folded into "workable" ones. Therefore, the quiet upside is that people don't actually lose their jobs as technology advances. It creates new jobs for those newly expanded works instead. Technology keeps inventing new goals, new standards, new things to want, and that plasticity of free will is what has been expanding the world all along. Then you might ask, are humans merely puppets? Maybe. But not entirely. Acceleration supplies the power to expand, not the direction. It is us who direct the expansion ourselves. That's why technology acceleration isn't a doomsday machine speeding the old world toward saturation but a way to reach the new one. To me that's the most humanist vision there is.

### 6

I strongly reject the idea that AI will eventually "learn" the taste. If AI could catch up that fast, and if taste mattered enough for anyone to chase it, then the big, rich, smart companies would already be shipping tasteful products. They mostly aren't. Outside a handful of firms run by people with real taste, most are still producing human slop. And taste keeps moving. Taste is an endless ritual and intention, not a finished state or goal. You can copy Liquid Glass, but you can't copy the first idea behind it, the reasoning, the discarded versions, the itch its makers felt while refining it, or their instinct for the next alternative. AI's taste will always be a fast follower.

### 7

I'm an aesthete. I've loved UI and UX for as long as I can remember, especially the ones with that satisfying, springy feel. On the web it's Vercel and Linear; on iOS it's Zenly 4.0, Airbnb, Luma, Bevel. I'm particularly fond of apps that follow Apple's Human Interface Guidelines closely.

### 8

I am an etymologist. Naming and coining them (or just deciding to) is work only a free spirit takes on, and I'm drawn to archaic words and made-up ones. Digging up an old word like seaflooding[^6] can make my whole day. I coin my own too: an acronym, [Doorknock](/E84BFC); an era notation, [誠鉉](/E95CAB); a neutral job notation for tech founders, [Enginifounder](/452EB4). The irony of naming everything is that I still don't have my [self-bestowed name](/8A1CBA), and the handle anaclumos was [pure accident](/A798A0).

[^6]: 개부심. a native Korean word for the rain that returns after a lull in the monsoon floods and rinses the leftover silt away; figuratively, to renew something completely

### 9

I don't particularly like the Republic of Korea, but I feel a huge responsibility toward my people and their Koreanness. Koreanness is full of wonders nobody really dug into, yet they still feel underdeveloped. That's because it never got ruminated enough. Take Greco-Roman culture: the actual BC version would look crude to modern eyes. It reads as high culture now only because sixteenth-century Florence spent centuries reworking those raw materials into something refined. Every culture starts out cringeworthy. Judging unruminated Koreanness by Hollywood's bar is like grading a 5yo's scribble against an art-school exams. Who knows that 5yo would grow up to be the next Picasso? I've named this reworking the [Neo-Korean Renaissance](/A39EC8).

### 10

I used to invest super seriously. I ran my [own quant engine](/BEC9CB), and in 2024 I briefly topped the weekly Toss trading leaderboard twice (Korea's Robinhood). These days the flood of information and the curse of knowledge have worn me down, so I just dollar-cost average into a few companies I like. A quirk: I also collect old currency and bonds. [Collecting seems to run in the family](/27E083). There's no grand plan. I just love holding a physical piece of someone's debt from back when they towered over everyone. My latest is a 1950s Soviet bond.

### 11

I think games might be the only way out of the concentration drought. Console games especially, the ones that ask for stamina and focus over long stretches, feel like literature to me. If the novel was the literature of the 19th century and film the literature of the 20th, [games are the literature of the 21st](/41F737). Really savoring a good one is both a skill and a privilege. My favorites: Ghost of Tsushima, Death Stranding 1 & 2, Detroit: Become Human, Splatoon 3, Super Mario Odyssey and Super Mario Bros. Wonder, SANABI, The Last of Us 1 (& 2, and honestly, 2 was fine for me), and Spider-Man 1 & 2. But the number one of my whole life is the Rhythm Heaven & Rhythm Tengoku series!

### 12

I aspire the new frontiers that computer science keeps opening. But I know the same technology can be turned to harm and eat away at what makes us human, and knowing that, I still fall into digital media from time to time. Some ask how a man who makes his living off computers can be wary of software, and the answer is that it took me years to separate "I like software" from "I like software, no matter what." Endorsing nuclear power isn't the same as endorsing the destructive fury of the atomic bomb. Even so, I've decided full independence from the internet is basically impossible for someone in this line of work, so I keep experimenting. In 2019 I carried a 3GS on iOS 6; in 2026, an SE1 on iOS 16; I've tried the Minimal Phone, the Sidephone, the Light Phone. Each one comes up short in its own way. A tool must disappear, but a dumbphone just adds one more. I still need to sit with that.

### 13

The world is short on selves. At fourteen I decided that individuality, creativity, and dreams had to be walled off and protected, and of the three, individuality is where everything starts. I prefer the word "color" to "individuality," and most of my color comes down to what I like. We all become what we like in the end. I treated it as survival: without individuality, the metaphor called "me" is as good as dead. I think I've mostly figured out who I am by now. But the world moved on. What matters isn't that information exists, it's that something real shows up in the world. So I started wishing[^7] this "me" could become something physical, something you could actually reach. Anyway, it's frighteningly easy to drift along with no resistance. You do things because others do, watch what others watch, do what others say you have to. It happens more than you'd think, and the second you ask "why," there's no answer. To find one you have to know what makes you you, and that you-ness is fainter than the sound of a midsummer breeze. So the self must be excavated[^8]. Individuality isn't handed to you. Only by noticing, recording, and studying yourself do you catch it: huh, so I like this? At some point I decided to invest in that without reservation, to express whatever I love intensely and without a shred of shame. The reward was my favorite line: "this is so you lmao."

[^7]: That wish, written down in 2021, is the site you're reading right now.

[^8]: A self forms when the shabbiest little anecdotes get rediscovered and deified.

### 14

My interest in accessibility and internationalization is close to an obsession, for two reasons. First, living in the US, I kept hitting walls whenever I tried to reach local culture and media. Watching brilliant Korean friends lose chances to a language barrier, I realized those barriers become a kind of discrimination whether anyone means them to or not. Second, a product that handles a11y and i18n well tends to be the more careful, thorough, well-built one, because nobody gets to that work until everything else has already cleared a high bar. It's like the fastest way to gauge a product's exquisiteness.

### 15

I'm a very extroverted person. Or I was. I still love talking with smart people, and I like being the least capable one in the room. Someday I want to build something that wins on real quality instead of loud marketing or bravado. That's why I keep my role models close: Jensen Huang's persistence, Lisa Su's turnaround leadership, and Guillermo Rauch's pioneering streak.

### 16

I want to be the Medici of science and technology. In the wreckage after the Black Death, as a little hope started to stir, the Medici backed brilliant people and seaflooded[^9] a whole industry of art. What moved me is the idea behind it: that you can move exceptional people to build a better future. So I decided to be one of those people and build it with them. The closest thing to the House of Medici today, I think, is a genuinely good company. I mean to turn my daydreams real, lay the road as I walk it, and grow an ecosystem where people can chase new adventures as far as they want. The most beautiful future blooms from present pain.

[^9]: The Renaissance was born from ashes, not from peace and prosperity. It's my favorite moment in all of history.

## Experience

- Search ML Engineer, TwelveLabs, Aug 2026 to Present
  Building the search infrastructure for multimodal embeddings of video, audio, and text at scale.
- AI Research Engineer, Lunit, Jan 2026 to Jul 2026
  Built a Slack AI agent and MCP server that automates internal research workflows.
- Research Engineer, Lunit, Jun 2024 to Dec 2025
  Moved AutoML platform to multi-cloud, built local GPU tracking SDKs, and cut all loadings to sub-1s.
- Software Engineer Intern, Grammarly, May 2023 to Aug 2023
  Shipped a configurable significance-threshold feature. Lifted powered experiments by 8%p.
- Software R&D Engineer Intern, Karrot, May 2022 to Aug 2022
  Prototyped mini-app runtimes with web-standards.
- Software Engineer Intern, Woowa, Jul 2020 to Aug 2020
  Built four demo services: auth, kanban, bookkeeping, e-commerce.

## Education

- M.S. in Computer Science, University of Southern California, May 2024
  GPA 3.80 / 4.00 (Prof. Jose-Luis Ambite)
- B.S. in Computer Science, University of Southern California, May 2023
  Boeing Scholarship, Magna Cum Laude, GPA 3.81 / 4.00
- International, Korean Minjok Leadership Academy, Feb 2020
  First Class Honor in Macroeconomics, ACT 35/36 (US Top 1%)

## Awards

- Bronze Medal, Korean Olympiad of Informatics, 2019
- WWDC Scholarship, Apple, 2018

## Elsewhere

- [GitHub anaclumos](https://github.com/anaclumos)
- [LinkedIn anaclumos](https://linkedin.com/in/anaclumos)
- [X @anaclumos](https://x.com/anaclumos)
- [Email hey@cho.sh](mailto:hey@cho.sh)

See also: [/mini](/mini), [/000000](/000000), [/llms.txt](/llms.txt).
</profile>

<mini_apps>
- Saturi, https://cho.sh/mini/saturi: Translate standard Korean into major Korean saturi styles and compare them side by side.
- Collections, https://cho.sh/mini/collections: Curated paths through the notes: themed groupings to read in order.
- Domains, https://cho.sh/mini/domains: Type a name and see if it's taken and any good: domains, handles, packages, GitHub, the dictionary, and an honest AI vibe score, all at once.
- Problems, https://cho.sh/mini/problems: Pick a problem, write your solution in the built-in editor, and the judge runs it against real test cases while you watch.
- Chess, https://cho.sh/mini/chess: Play a quick chess game against Sunghyun or AI, keep AI beside you as an advisor, or play a friend on two phones.
- Sans, https://cho.sh/mini/sans: A rounded typeface supporting 9 weights across Latin, Cyrillic, Greek, Vietnamese, Korean, Japanese, and Chinese.
- Token Usage, https://cho.sh/mini/usage: How much I spend on my AI coding tools, day by day.
- Tokenmaxxing, https://cho.sh/mini/tokenmaxxing: Live quota headroom across my pooled Claude Code, Codex, and Grok accounts, pushed from my workstation every 10 minutes.
- Mock Interviewer, https://cho.sh/mini/mock-interviewer: Upload what you want to be quizzed on, then run a live voice mock interview grounded in your documents.
- Dice, https://cho.sh/mini/dice: Roll a handful of 3D dice with real physics and add up the result.
- Coin, https://cho.sh/mini/coin: Flip a 3D coin with real physics. Heads or tails, your call what they mean.
- Russian Roulette, https://cho.sh/mini/roulette: Spin a toy revolver to draw a name. Last one standing wins.
- Bottle, https://cho.sh/mini/spinner: Spin a 3D bottle to pick someone from the circle.
- Health, https://cho.sh/mini/health: Sleep, readiness, activity, steps, and heart data from my health devices.
- Jukebox, https://cho.sh/mini/jukebox: Browse Sunghyun Cho's recent Apple Music rotation. Pick a record, play a preview, and find your next favorite song.
- Liar Game, https://cho.sh/mini/liar: One phone, everyone plays: all but the liar see the secret word. Describe it, vote, and catch the liar before they bluff through.
- Mafia, https://cho.sh/mini/mafia: The classic party game of Mafia with phone-based player rooms and a live AI narrator on the host device.
- Apple Wallet Card, https://cho.sh/mini/wallet: Create Sunghyun's Apple Wallet business card with the place and date you met.
- Papers, https://cho.sh/mini/papers: Paste an arXiv link or upload any paper's PDF and reading agents unroll it into a readable page: margin notes where the text gets hard, a guide for every figure, and an adversarial review of what the authors left unstated.
- News, https://cho.sh/mini/news: AI news alerts: pick a topic and AI fetches, scores, and summarizes the stories worth reading.
- Rubber Duck, https://cho.sh/mini/rubber-duck: Pick any topic and talk it through out loud with a voice duck that listens, asks sharp questions, and helps you untangle your own thinking.
- Salpakan, https://cho.sh/mini/salpakan: Play Salpakan, the Filipino Game of the Generals: hide your ranks, march your flag, and outguess AI or a friend.
- Blackjack, https://cho.sh/mini/blackjack: A casual blackjack table on cho.sh: seat your friends or deal AI in, and play against the house dealer.
- Dalmuti, https://cho.sh/mini/dalmuti: The climbing card game of medieval social mobility on cho.sh: gather four to eight players, or seat a few AI, and race to shed your hand first.
</mini_apps>

<public_repositories>
</public_repositories>

<access>
Use whichever of these your environment supports. Every one is read-only and needs no sign-in.

- MCP: connect to https://cho.sh/mcp (Streamable HTTP). Tools: search_cho_sh (full-text search over the notes, mini-apps, problems, news and papers), get_page (a page as Markdown), list_surfaces. Server card: https://cho.sh/.well-known/mcp/server-card.json.
- HTTP and CLI: `curl -H 'Accept: text/markdown' https://cho.sh/<path>` returns any page as Markdown. A .md or .json suffix does the same (https://cho.sh/about.md, https://cho.sh/mini/health.json). `curl 'https://cho.sh/api/search?q=<query>&limit=20'` searches the same index as the site. https://cho.sh/llms.txt lists the pages and https://cho.sh/openapi.json describes the endpoints.
- Notes: a note is the 6-character hex path (https://cho.sh/<hex>.md). Search first, then read the best matches in full.
- GitHub: `gh repo list anaclumos --visibility public`, `gh repo view anaclumos/<repo>`, or the repository pages themselves.
- Web: search the public web for "Sunghyun Cho" and "anaclumos", and read https://github.com/anaclumos, https://linkedin.com/in/anaclumos, https://x.com/anaclumos.
</access>

<method>
1. Read my project below. List its topics, its stage and the decision I seem to be facing.
2. Fetch the home page and the about page, then run 3 to 6 searches built from my project's topics. Read the 2 to 4 best notes in full.
3. Check the mini-apps and the public repositories for work that overlaps with my project, and open the closest ones.
4. Search the public web for recent public material by or about Sunghyun that fits my project.
5. If you cannot call any of these, say so and work from this prompt only.
</method>

<answer>
Write the answer in the language of my project, in this order:

1. Fit: one paragraph on where Sunghyun's work overlaps with my project. If the overlap is small, say so.
2. What he has already thought about: 3 to 6 items. Each item links one specific note, mini-app, repository or public page and says in one sentence how it applies to my project.
3. Where he can help: concrete offers in order of value, such as a review, a prototype, an introduction or a decision to talk through. Each offer names the evidence it rests on.
4. Questions he would ask me before starting.
5. A message to him of at most 120 words that I can send to hey@cho.sh.

Rules: state only what a source you read supports, and link that source. Mark an inference as an inference. Do not invent notes, projects, employers, dates or links. Do not flatter.
</answer>

## My project

Paste your project intro, a link to a repository, an old doc or rough notes here, then send this prompt.