[Noe Valley Coffee Roasters — January 2014, 2:15 PM]
Sarah was already at the corner table when Ethan arrived, the worn notebook open beside a half-eaten muffin and a black coffee she'd clearly made herself behind the counter before clocking out. She was writing — fast, small handwriting, filling a page with diagrams and annotations. She closed the notebook when she saw him, but slower than last time. The guard was lowering by degrees.
"Your training run," she said, instead of hello. "Did it converge?"
Ethan dropped into the chair across from her. He'd been up since six, watching the loss curve on ChronoCloud's dashboard like a heart monitor in an ICU. "Loss hit 3.4 at epoch forty-two. Still dropping. The model's generating coherent fragments — not sentences, but phrases that hang together. Token predictions are non-random."
"On how much data?"
"Thirty million tokens. Project Gutenberg, cleaned and tokenized."
Sarah's eyebrows went up half an inch. "Thirty million tokens through a novel attention architecture, and you're getting coherent fragments after forty-two epochs? On what hardware?"
"Cloud provider. Proprietary."
"Which one?"
"Can't say."
The eyebrows came back down. The guard went back up. Sarah picked up her muffin and took a bite, chewing slowly, watching him with the evaluative patience of someone deciding whether to invest more time in a conversation that kept hitting walls.
"Here's the thing," she said. "You've shown me code that uses an architecture I've never seen published. You have access to compute that you won't name. You're offering me a job at a company that — I looked you up — has one client paying five hundred dollars a month for a dashboard nobody uses. And you want me to help you pitch VCs on technology that the entire machine learning community would call science fiction."
"That's accurate."
"So either you're a genius, a con artist, or something I haven't considered."
"I'm an engineer who has access to something other people don't. I can't explain how. The technology is real."
Sarah set down the muffin. "That's not an answer."
"I know."
A long pause. The café's acoustic playlist cycled to a new song. Someone at a nearby table laughed at something on their phone.
"Show me the pitch deck," Sarah said.
Ethan pulled the laptop from his bag and opened the latest version — fourteen slides, the consumer-friendly revision he'd built at two in the morning. He turned the screen to face her.
Sarah read. Slowly. Each slide got a minimum of thirty seconds, which was twenty-five seconds longer than most VCs would give it. Her expression cycled through phases: interest, confusion, skepticism, something that might have been reluctant recognition.
"Slide four," she said. "You're describing the architecture as 'a novel neural network approach to language generation.' That's technically accurate and completely useless. What does it do?"
"Generates text."
"So does a Markov chain. What makes this different?"
"Coherence. Context. The attention mechanism lets the model consider the entire input simultaneously instead of processing it sequentially. It understands relationships between words regardless of distance."
"Then say that. 'Our AI reads entire documents at once and writes responses that understand context.' Don't say 'novel neural network approach.' Nobody outside a research lab knows what that means, and nobody inside one will take you seriously without a paper."
Ethan opened the slide editor. Sarah leaned over and started typing. Her fingers moved with the same purposeful speed he'd seen during the bug catch — precise, unhesitating, each keystroke deliberate.
"Slide six is worse," she continued, deleting three bullet points and rewriting them. "You're explaining attention heads. VCs don't care about attention heads. They care about output. Can it write marketing copy? Can it summarize documents? Can it answer customer service emails? Lead with what it does. Bury how it works."
"But the architecture is the innovation—"
"The architecture is your innovation. The VC's innovation is writing a check. Those are different conversations." She grabbed the laptop fully, rotating it to face herself, and began restructuring the slide order. "Put the demo slide first. Then market size. Then team. Architecture goes in the appendix where nobody looks unless they're already interested."
Ethan watched her work. The pitch deck that had taken him six hours to build was being disassembled and rebuilt in fifteen minutes. Sarah moved slides with the casual confidence of someone who'd seen a hundred pitch decks — not as a founder, but as someone who could see through the presentation to the product underneath and knew exactly where the gap between them lived.
"You don't have a demo slide," she observed.
"I don't have a demo."
"Then you don't have a pitch. You have a theory with slides."
"The model's still training—"
"When does it finish?"
Ethan checked the time. The training run had been going for sixteen hours. ChronoCloud's dashboard showed loss at 2.8 and still dropping. "Tonight. Maybe tomorrow morning."
"Can it generate anything a person would look at and say 'a computer did this?'"
"Fragments. Short sequences. The model's too small for paragraphs."
"Then your demo is fragments. Show a VC a computer writing three coherent sentences about something it wasn't trained on, and they'll be more impressed than any slide deck." Sarah pushed the laptop back toward him. "Here. I reorganized. It's not good — it's less terrible. But without a demo, you're selling smoke."
Ethan scrolled through the revised deck. She'd cut it from fourteen slides to nine. The architecture section was gone entirely, replaced by three case studies — hypothetical, but grounded in real business applications. Automated content generation for marketing agencies. Document summarization for legal firms. Customer service automation for e-commerce. Each one had a market size figure that Sarah had researched on her phone while Ethan watched.
"When did you learn market sizing?" he asked.
"I read." She said it the way she said everything — flat, factual, leaving no room for follow-up. Then, softer: "My startup. Before it failed. I was the only person who could do both the technical and the pitch. The CEO couldn't explain what we built, and the CTO couldn't explain why anyone would want it. So I learned."
---
[Same Café — One Hour Later]
They'd ordered more food. Ethan's half of the table held the laptop, a turkey sandwich he'd bought for $8.50 and defended against Sarah's casual grazing, and a second latte. Sarah's half held her notebook — still closed — another black coffee, and a plate of fries that she ate one at a time while talking.
"The problem isn't your technology," Sarah said, dragging a fry through ketchup. "The problem is the year. If this were 2016, you'd be pitching into an AI boom. VCs would be begging to write checks. But it's 2014. Deep learning is an academic curiosity. The VCs funding AI right now are academics themselves, or contrarians betting against the market. Mainstream Sand Hill Road doesn't want this."
"So who does?"
"Contrarians. People who fund things that sound impossible because they know the possible stuff is crowded." Sarah ate the fry. "Do you have any meetings?"
"One. David Park at Basecamp Ventures. He's an associate. Not a decision-maker."
"An associate at a small fund is better than a partner at a big one. Partners at big funds need consensus. Associates at small funds need a win. If David brings his partners a deal that returns thirty X, he gets promoted. That's a hungry person."
The analysis was sharp. Ethan hadn't thought about the VC structure that way — he'd been focused on the technology, not the incentives of the person across the table.
"First pitch meeting is in two days," he said. "David replied this morning. Coffee at Philz on Wednesday."
"Philz is good. Casual. Low stakes." Sarah picked up another fry. "What's your ask?"
"Seed round. Two hundred fifty thousand at a five million pre-money valuation."
"Too high. You have no traction, no team, and no demo. Ask for a hundred fifty at three million. It's low enough that David can make the case without sticking his neck out, and high enough that you're not giving away the company."
"You're helping a lot for someone who hasn't taken the job."
Sarah chewed. Swallowed. Met his eyes directly for the first time since the conversation started.
"I've been making lattes for six months. I have a notebook full of architectures that nobody will ever see because I can't afford compute and I don't have credentials. I left Stanford because the system decided my funding was less important than a tenured professor's sabbatical. The last startup I joined promised equity and delivered layoffs. So when someone walks into my coffee shop with code I've never seen before, using techniques that don't appear in any paper I've read — and I've read all of them — and tells me he's building something real? I don't say no immediately. I evaluate."
She picked up the last fry.
"I'm still evaluating."
Ethan nodded. He didn't push. Didn't sell. The instinct from his previous life — the ML engineer who'd never managed people, never recruited, never needed to convince anyone of anything more consequential than a code review — told him to argue. To list reasons. To close the deal.
But something else — something he was beginning to trust, an intuition that existed outside his four supernatural abilities — told him that Sarah Chen didn't respond to sales pitches. She responded to proof. Evidence. Working code and decreasing loss curves and the quiet confidence of someone who knew what they were building because they could see it.
"The model finishes training tonight," he said. "Come to the apartment tomorrow. I'll show you the output."
Sarah looked at the pitch deck on his screen. Nine slides. A company that didn't exist yet. A technology that shouldn't be possible. And a founder who couldn't explain where any of it came from.
"I still don't fully understand what you're building," she said. "But whatever it is, it's not fake."
She picked up her notebook. Tucked it under her arm. Stood.
"Send me the address. I'll come after my shift." She paused at the edge of the table. "And buy groceries. Your apartment probably looks like a server room and smells like ramen. I'm not debugging code in a biohazard."
The door chimed behind her. Ethan sat alone with the pitch deck and the cold remains of his sandwich, which Sarah had somehow reduced to a heel of bread and a single lettuce leaf during the course of the conversation.
His phone buzzed. David Park.
Wednesday at 2 PM works. Philz on 24th. Looking forward to hearing what you've been up to.
The first pitch meeting. Forty-eight hours away. A nine-slide deck, a model that was still training, a bank account draining by the hour, and a potential CTO who hadn't said yes but had bought him three hours of her time and reorganized his entire strategy.
Ethan signaled the barista — not Sarah's replacement, a teenager who looked bored enough to be dangerous — and ordered another latte. He needed the caffeine. The next forty-eight hours required more code, more preparation, and the particular kind of focus that existed in the narrow space between panic and determination.
The latte arrived. Mediocre. Not Sarah's work. He drank it anyway and opened the laptop.
The ChronoCloud dashboard showed loss at 2.6. Dropping. The model was learning.
Now he needed to learn something harder: how to sell the future to a man who lived in the present, in a room that smelled like artisanal coffee, with nothing but nine slides and the absolute certainty that the math worked.
Author's Note / Promotion: Your Reviews and Power Stones are the best way to show support. They help me know what you're enjoying and bring in new readers! You don't have to. Get instant access to more content by supporting me on Patreon. I have three options so you can pick how far ahead you want to be: 🪙 Silver Tier ($6): Read 10 chapters ahead of the public site. 👑 Gold Tier ($9): Get 15-20 chapters ahead of the public site. 💎 Platinum Tier ($15): The ultimate experience. Get new chapters the second I finish them . No waiting for weekly drops, just pure, instant access. Your support helps me write more . 👉 Find it all at patreon.com/fanficwriter1