7,000 ACCOUNTS · 16× REVENUE · 29 BUILT
1,200 accounts to 7,000. Then I built the rest alone.
Find where it jams, and turn that spot into a system.
Taerim Hwang — Software is my trunk. I dropped out of school at seventeen and started with software and security in the IT Bank security course. From there came the branches: security installations, Facebook social growth and advertising, a bag distribution company (entry-level to head of strategy in a year, a board seat offered in the second), then wholesale distribution and large-scale device operations. Today I turn the problems I met on the ground into software, AI and automation.
I have been teaching myself since seventeen.
Software is the trunk. At seventeen I dropped out of school and started learning software and security in the IT Bank security course (a white-hat hacker programme), and I have not let go of it since. Everything else grew from that trunk as a branch. I handled access control and CCTV hardware in the field, learned where attention moves from social media and advertising, and saw how goods turn into money in wholesale distribution. My family ran a garment factory, so manufacturing was familiar from early on. None of it is certified by a diploma. All of it I actually ran.
I think of it as a skill tree. What you take determines what you can take next. The breadth is not ambition — it is that once the pieces connect, one person can run many things at once. You have to know how to sell to know what to build, know how to build to automate the repetition, and know how to automate to run several things alone.
Making the product
Deciding what to sell, sourcing or manufacturing it, and turning it into something real. Planning, sampling, factories, lead times, cost.
Making it sell
Wrapping it in a brand, telling people about it with content, and building the path from attention to purchase — domestically and country by country.
Building the sales structure
Wholesale and retail, B2B and B2C, inventory, orders, permissions. Not one sale, but a structure that keeps running.
Automating that structure
Moving what people repeat into systems. Building it myself, and using AI to change the scale one person can handle.
Started with software — dropped out at 17, IT Bank security course (white-hat hacker)
Age 17–18School did not teach what I wanted to make. At seventeen I dropped out and joined the IT Bank security course, learning software and security. Every job after that grew as a branch of this trunk, and I have not stopped since.
Security systems — access control and CCTV
Age 19–21Installed access control and surveillance equipment in the field, wired it, connected the network and made it run. Hardware products do not end at the sale — installation and operation follow.
Social growth and performance advertising
Age 23–26Ran Facebook groups and pages — the largest group at 446,000 members, roughly two million across all of them, and eleven pages including one at 830,000 likes. Grew a YouTube channel to 103,342 subscribers and 24 million lifetime views with monetisation approved. Facebook influence ranked 20th nationally, top 1%.
The crypto bet
Age 27I consolidated what I had built and put all of it in one place without verifying my conviction. I lost heavily. The problem was not that the conviction was wrong — it was that I had no mechanism to test it. Two habits stayed: refute my own conclusions with data first, and put a loss ceiling into whatever I build.
Bag distribution company — entry-level to head of strategy in one year
Two yearsRetail accounts grew from 1,200 to 7,000 and annual revenue went from around one billion to sixteen billion won. I widened the channels three ways: live commerce with hosts I booked, outbound B2B calling from contact lists I gathered myself, and Python automation that listed and managed products across roughly 230 sales channels. There was no AI then, so I wrote it by hand. I was offered a board seat and declined it — I wanted to run my own business.
Fashion B2B distribution
Age 27–29Handled women's apparel, bags and accessories, and menswear across different wholesale markets. Learned concepts you cannot get from documents — goods owed, samples out, on hold, pending.
Online commerce — multi-category distribution, branding, manufacturing
Bags and accessories, sunglasses, cosmetics, pet food and treats, furniture and appliances. Planning, sourcing from China or manufacturing, branding, imagery, detail pages, SEO, listing, selling and support — all of it myself. Also sold through overseas marketplaces: Shopee across nine countries, Lazada across five, Rakuten, Qoo10, Yahoo Shopping, Amazon US.
Large-scale device operations
Age 29–34Built and ran an environment of 40+ phones and 8 PCs alone, on a purpose-built rack with USB hubs, power distribution and active cooling. Ran twelve clients on a single machine, 96 sessions across eight. Distributed devices across major cities and managed them remotely.
Building — 29 programs
Since May 2026Trading intelligence with two years of walk-forward backtesting, a multi-tenant content SaaS, an SMM platform, a wholesale POS, AI seller automation, personalised AI services. 121,382 lines, alone, with AI agents as a constant development partner.
I never intended to commercialise all of it. One is closed through payment and live; the rest are tools for my own work or things I built to confirm I could. This is not a list of finished products — it is a record of what I can make.
Wholesale POS
Mode-switching commerce system for wholesale and retail
Wholesale markets open at 8pm and close at 5am, so calendar dates split a single day of trading in two. A slip holds more than sales — goods owed, samples lent, items pending, items on hold. I made 'business day' a first-class concept and gave six line types their own effect on inventory and totals. Permissions are not hidden in the UI; the backend actually blocks them.
Sovereign Apex
Proprietary trading system with two years of strategy research
Strategies always look good in a backtest. I extended the sample to two years and validated first and second halves separately as a walk-forward. The 80% win rate I was aiming for turned out to be an artefact of a small sample, so I discarded the target. When the data disagrees with the conclusion you wanted, you throw away the conclusion.
PickWise
AI content automation SaaS
Generation is one API call. Stopping a false sentence from being published is the actual design problem. I chained generation, verification, SEO optimisation and scheduled publishing into one pipeline with guardrails at each step, and used a bring-your-own-key model so cost risk never accumulates on the platform. Twenty-one blogs run on it.
Wolha
Korean fortune-telling engine with AI interpretation
I built the calendar calculation engine myself to remove external dependencies, cached AI interpretation to cut cost, and added a rule-based fallback so the service still answers when the model fails. Prices are decided from a server catalogue and checked against the order right before approval — the client is never trusted with an amount.
Nineteen more: an SMM platform running live, AI seller automation, real-time market data pipelines, a decision dashboard, chart analysis, audience analytics, legal case tooling, a seller AI studio, a nine-service control panel, and others.
Software
Backend, frontend, data and deployment, handled by one person. All 29 programs came from here.
AI product design
Putting AI into a service pipeline, not a demo. Verification, fallback and cost are part of the design.
Working with AI
AI is used well by people who have done the work. Building 29 programs since May came from this.
Automation and vision
Finding what people repeat and moving it into systems. Screen capture, recognition, browser automation.
Trading and data
Real-time ingestion through indicator computation and two years of backtest validation. Thirteen custom indicators written from scratch.
Commerce automation
Thirteen steps a seller repeats — discovery to listing — connected into one workflow.
Commerce and distribution
Apparel, bags, sunglasses, cosmetics, pet food, furniture, appliances. Wholesale and retail, B2B and B2C, domestic and overseas.
Manufacturing
Deciding what to make and connecting it so it gets made. Planning through production to distribution, without a break.
Branding
Making a product and making it a brand people buy are different jobs. Name, line-up, detail pages, channels.
Marketing and growth
Traffic to interest to conversion to sale. Search, social, performance advertising, mass blog operations.
Large-scale operations
Around two million across communities, 40+ phones, 8 PCs, devices distributed across cities and managed remotely.
Global marketing
Japan, China and Southeast Asia treated as separate markets. Not prepared for — actually sold in.
Video and content
Planning, editing and motion graphics. A channel at 103,342 subscribers and 24 million views.
Hardware and systems
PC builds and optimisation pushed to their limit. Device farm racks, capture and audio systems.
Product and system design
Starting from the problem rather than the feature, and breaking business processes into systems.
Business experience
Growing a company from the inside, and running one from the outside. Both.
Connecting the fields
Not the best in any single field. But able to join manufacturing, distribution, branding, marketing, content, IT and AI into one business.
These are counted, not claimed. Every figure was measured directly, and how it was measured is written beside it. Lines of code do not prove skill — but they do say what the months since May were spent on.
- Retail account growth and revenue figures are the company's results while I led strategy, not my personal performance converted into numbers.
- Backtest returns are simulation results, not realised profit.
- One service is closed through live payment. The others were never intended for commercialisation — they are tools or proofs.
- Android builds are debug builds, not store releases.
- Some earlier roles have no surviving record of exact dates, so no dates are given. I did not fill them in by guessing.
- Source code is not public. Most of it is either in operation or used in an actual business. I am happy to walk through the code and architecture directly.
Collaboration or commission, either works. An idea with nothing built yet is fine — deciding what to make is part of the job.