Chapter 22: The Stall
The learning module went live on November 28th, and Bridge learned its first lesson at 9:07 AM from a second-year biology student who corrected a task classification by dragging an email about a study group from “Lab Report” to “Exam Prep.”
It was a small correction. One email, one drag, one click. But inside Bridge’s code, that single user action triggered a cascade: the vector embedding engine compressed the correction into a 64-dimensional numerical representation, stored it in the lightweight database Dojun had spent three weeks building, and updated the classification model’s weights by a fraction of a percent.
Bridge had learned. For the first time, it had observed a human’s preference and adjusted itself to match.
By the end of the first day, 847 corrections had been logged. By the end of the first week, 6,200. By December 1st, they had surpassed the 10,000-example threshold, and the model’s accuracy had climbed from 87% to 91%.
“Four percent in three days,” Minjae said, staring at the dashboard. “The model is learning faster than I did in my algorithms class.”
“The model has better training data than your algorithms class,” Dojun said. “Real humans making real corrections on real tasks. That’s the cleanest signal you can get.”
“And the memory issue?”
“Solved. The vector embeddings are fifty times smaller than the raw correction data. We can store a million corrections in the same memory budget that used to overflow at a hundred.” He allowed himself a moment of satisfaction. The compressed representation was elegant—not the sophisticated neural embeddings of his previous life, but a simpler scheme using locality-sensitive hashing that achieved 90% of the quality at 1% of the computational cost.
“Users are noticing,” Hana reported from her desk, scrolling through the feedback channel. “Listen to this: ‘Bridge used to put my tutoring emails in the wrong group. I corrected it twice, and now it gets it right every time. It’s like training a very smart puppy.’ We’re getting a lot of puppy metaphors.”
“Puppies are good,” Minjae said. “People love puppies. Our marketing slogan: ‘Bridge — Your Digital Puppy.'”
“We are absolutely not calling our enterprise software a digital puppy,” Hana said. But she was laughing.
The learning module’s success rippled outward. Daily active users, which had plateaued at 312 in October, jumped to 380 as word spread through campus that Bridge was “getting smarter.” The waitlist—which Hana had set up on the new website—reached 1,200 names within a week of launch. Students from Korea University and Yonsei were asking if they could join.
“We need to expand beyond SNU,” Hana said during the Monday standup. “The demand is there. If we limit ourselves to one campus, we’re leaving growth on the table.”
“Expanding means more servers, more support, more infrastructure,” Dojun cautioned. “We have three people.”
“We have fifty million won. Choi Eunji’s investment was for exactly this—scaling. We hire one engineer and one support person, expand to three campuses, and aim for five thousand users by spring.”
“Five thousand.” Dojun ran the numbers in his head. Server costs, bandwidth, support load. It was feasible—tight, but feasible. “We’d need to optimize the data pipeline. Right now it’s designed for five hundred concurrent users. Five thousand requires a complete rewrite of the sync layer.”
“How long?”
“Two months. If we hire an engineer who can handle the backend while I focus on the learning algorithm.”
“Then let’s hire.” Hana’s pen tapped the whiteboard. “I’ll draft the job posting. Minjae, you set up the interview process. Dojun, you write the technical assessment.”
“We’re hiring our first employee,” Minjae said, sounding dazed. “We’re a real company that hires people.”
“We’ve been a real company since September,” Hana reminded him.
“Yes, but hiring makes it feel real. Before, we were three friends in a closet. Now we’re three friends in a closet looking for a fourth friend who’s willing to work in a closet.”
“That’s going on the job posting,” Hana said. “Working environment: intimate. Team culture: closet-positive.”
December arrived with the sudden, bone-deep cold that turned Seoul into a city of scarves and steam. The campus was quieter—students retreated to libraries and heated cafes, leaving the walkways to the wind and the occasional brave jogger.
Dojun still ran every morning. The cold didn’t bother him—or rather, it bothered him in a way that felt like a privilege. In his previous life, the last winter he had experienced was through a hospital window, watching snow fall on the Han River while machines beeped and morphine dripped. This cold—sharp, real, stinging his lungs and reddening his cheeks—was the cold of being alive.
The semester’s end brought a cascade of deadlines: final exams, the KETI research report, Bridge’s expansion plan, and the Seokho collaboration’s first milestone. Dojun managed them the way he managed everything now—with Hana’s color-coded schedule pinned to his wall and his mother’s voice in his head saying “do the right things first.”
Final exams were straightforward. He got A-pluses in everything except Korean History, where he received an A for an essay arguing that Joseon-era signal fire networks were “an early implementation of distributed message passing.” The professor’s comment: “Creative but anachronistic. This is a history class, not a computer science seminar.”
The KETI research report was completed in collaboration with Kim Taesik—a clean technical document showing a 17% reduction in power consumption during cache miss events. Kim submitted it to the Korean Electronics Technology Institute with a recommendation for continued funding.
“The KETI people are impressed,” Kim told him. “They want to fund phase two next year. That’s another year of research assistantship—with a stipend increase.”
“I appreciate it, Professor. But I might not be able to commit the same hours. Bridge is demanding more time.”
“I know. Bridge should demand more time—it’s a real product with real users. But don’t abandon the research entirely. The academic credibility supports everything else you do.” He paused. “And between us—the KETI work is where you learn discipline. Bridge is where you learn ambition. You need both.”
“Discipline and ambition.”
“The two ingredients of a career that lasts. Most people have one or the other. You have both, but they’re not in balance yet.” He sipped his terrible coffee. “They will be.”
The Saturday before Christmas, Dojun brought something to the market that he had been working on for three weeks in secret.
His mother was setting up the stall when he arrived, arranging the day’s banchan with her customary precision. She looked up and smiled—the automatic, warming smile she reserved exclusively for him.
“Early today,” she said. “Did you eat?”
“I ate.” He set his backpack on the counter and pulled out a laptop—not his Compaq, but a refurbished IBM ThinkPad he had bought from a graduating senior for seventy thousand won. “Mom, I want to show you something.”
“If it’s another computer program, I don’t—”
“It’s a computer program for you.”
She stopped arranging banchan. “For me?”
He opened the laptop and turned it toward her. On the screen was a simple interface—large buttons, clear Korean text, warm colors that Hana had chosen specifically because “ajumma eyes need contrast, not subtlety.”
“This is Bridge Market Edition,” he said. “I built it for banchan stalls. Specifically, for yours.”
The interface showed three sections: Inventory (“What do I have?”), Orders (“Who wants what?”), and Accounting (“How much did I make?”). Each section was a single screen with large, touchable buttons—though the ThinkPad didn’t have a touchscreen, so they used a mouse. But the design was deliberately simple: no menus, no dropdowns, no technical language. Just questions and answers.
“I put in your current inventory,” Dojun said, clicking on Inventory. A list appeared: kimchi (12 containers), japchae (8), kongnamul (15), kkakdugi (10), doraji (6). “When you sell a container, you press the minus button. When you restock, you press the plus button. At the end of the day, it tells you what to prepare for tomorrow.”
His mother stared at the screen. She didn’t touch the mouse. She didn’t speak.
“And here,” he continued, clicking on Accounting. “You enter today’s sales at closing. How many containers of each, at what price. It calculates your daily revenue, your weekly total, and your profit after ingredient costs.” He had pre-loaded the ingredient costs based on what he’d observed during his Saturday visits—the wholesale prices she paid for vegetables, noodles, sesame oil, containers.
“You know my ingredient costs?” she said.
“I’ve been watching you buy supplies for six months. Mrs. Kang charges you three thousand won per kilo for radishes. Mr. Hwang gives you sesame oil at twenty thousand per bottle if you buy three or more.”
“You’ve been taking notes.”
“I’m an engineer. We take notes.”
She was quiet for a long time. She reached out and touched the screen—gently, tentatively, the way you touch something you’re not sure is real.
“You built this for me,” she said. Not a question.
“You told me once that you keep everything in your head. Thirty years of inventory, customers, suppliers. I thought—” His voice caught. “I thought maybe you shouldn’t have to carry all of that alone.”
“I don’t mind carrying it.”
“I know. But you shouldn’t have to. That’s the whole point of what I do, Mom. Building things that help people carry less. And you’re the person I want to help most.”
She was crying. Quietly, without drama, the way she did everything—practically, efficiently, with minimum fuss. She wiped her eyes with her apron and looked at the screen again.
“Show me the minus button,” she said.
He showed her. She pressed it. The kimchi count went from 12 to 11.
“I sold one container of kimchi this morning,” she said. “To Mrs. Choi. She comes every Saturday at 7 AM.”
“Then Mrs. Choi is Bridge Market Edition’s first user.”
“Mrs. Choi doesn’t know she’s using a computer program. She just bought kimchi.”
“That’s the idea, Mom. The best technology is the kind you don’t notice.”
She looked at him. Not with the usual maternal scrutiny or the proud-but-confused expression she wore when he talked about algorithms. With something else entirely. Recognition. Understanding. The look of a woman who had spent thirty years running a business from memory and was seeing, for the first time, what her son’s life was actually about.
“Invisible technology,” she said.
“You remembered.”
“I remember everything my son tells me. Even the parts I don’t understand.” She pressed the minus button again. Kimchi: 10. “Mrs. Park bought two containers. She’s hosting a family dinner tomorrow.”
“You’re a natural.”
“I’ve been pressing buttons longer than you’ve been alive. Just not computer buttons.” She set down the mouse and took his hand. “Dojun-ah. This is—” She stopped. Started again. “When you were little, you told me you wanted to build things. I thought you meant buildings. Like your grandfather—he was a construction worker. I thought maybe you’d build apartments or bridges.”
“I build a different kind of bridge.”
“I know. And I didn’t understand for a long time. All these computers and programs and algorithms—I couldn’t see what they had to do with real life. With the market, with banchan, with the things that matter.” She squeezed his hand. “Now I see. This little program—this is what you mean. This is what it’s for.”
“This is what it’s for.”
“Then keep doing it. Keep building bridges.” She let go of his hand and turned back to the stall, wiping her eyes one more time. “Now help me restock the japchae. The containers go in the front. Not the back. The front, Dojun—customers eat with their eyes.”
“I know, Mom.”
“Then why do you always put them in the back?”
“Force of habit.”
“Habits can be changed. Even for engineers.”
He restocked the japchae—in the front—and spent the rest of the morning helping his mother learn Bridge Market Edition. She struggled with the mouse (“This thing moves like a drunk squirrel”) and accidentally deleted the entire doraji inventory (“Where did my doraji go? Did the computer eat it?”). But by noon, she had logged four sales, updated her inventory, and produced a daily revenue calculation that was, she admitted grudgingly, “probably more accurate than my head math.”
“Probably?” Dojun said.
“My head math has thirty years of experience. Your computer has one morning. I’ll reserve judgment.” But she was smiling—the particular smile of someone who had just discovered that the future was less frightening than she thought. “Come back next Saturday. I’ll have a list of things that need fixing.”
“A bug report?”
“I don’t know what that means. I have complaints. Same thing?”
“Exactly the same thing.”
Christmas Eve. Seoul glittered with lights—the secular, commercial Korean Christmas that had more to do with couples and cake than with religion. The campus was empty, the dormitories quiet, the Innovation Center closed for the holiday.
Dojun spent the evening in his apartment, alone by choice. Hana had gone to Busan to visit her parents. Minjae was with his family in Incheon. Seokho, predictably, was in his KAIST lab working on the collaboration paper (“Holidays are social constructs. Algorithms don’t take days off.”).
He sat at his desk with a cup of tea—barley tea, made from a packet his mother had pressed into his hands last Saturday—and opened his journal.
December 24, 2006. Nine months since I woke up.
Inventory of a second life:
One funded startup. Five hundred users, growing. A learning algorithm that works. An office with a wobbly chair that we still haven’t replaced.
One published paper (ISCA, June). One collaboration in progress (SIGCOMM, with Seokho). One research assistantship (KETI, phase two funded).
One mentor who knows I’m impossible and believes in me anyway. One rival who became a collaborator. One partner who is becoming the most important person in my life.
One mother who learned to use a mouse today and called it a “drunk squirrel.” She logged her banchan sales on a laptop and said, “Now I see what it’s for.” Nine months of work, and that sentence is the best review I’ve ever received.
This is not what I expected. When I woke up in March, I had a plan—build Prometheus Labs again, earlier, better. Avoid the mistakes. Change the future.
But the future I’m building doesn’t look like Prometheus Labs. It looks like Bridge. Smaller, quieter, more human. A product that helps five hundred students organize their homework, and a market version that helps one ajumma track her kimchi. Not the kind of thing that makes magazine covers. The kind of thing that makes lives a little easier.
Maybe that’s enough. Maybe that’s better.
I came back to build the future. But the future I actually want is simpler than I imagined: a mother who’s proud, a partner who trusts me, friends who show up, and work that helps the people I love carry less.
The web holds. The code runs. The people are here.
Merry Christmas, old man. You’re doing okay.
He closed the journal, finished his tea, and went to bed.
Outside, Seoul’s Christmas lights blinked in the cold—red, green, gold, a constellation of small human joys strung across a city of ten million. Somewhere, couples shared cake. Somewhere, families gathered around tables. Somewhere, a twenty-year-old man with sixty-three years of memories lay in a studio apartment and listened to the city breathe and felt, for the first time, that both of his lives—the one he had lost and the one he was building—had led him to exactly where he needed to be.
The ceiling stain, shaped like a lopsided heart, watched over him as he slept.
Tomorrow would be Christmas. And then a new year. And then the spring, and San Diego, and whatever came after.
But tonight, the web held. And that was enough.