Soo Youn (Julie) Park, Frontier AI, across borders and across the stack

00. About

Zero to one, across borders

Over seven years at SK Telecom, I built zero to one across borders, from the industry’s first telco-specific LLM and a five-party AI joint venture to computer vision and gigawatt-scale AI infrastructure. I’m now an MBA candidate at Harvard Business School, Class of 2028.

I’m especially interested in how the AI stack is evolving end to end: how inference gets cheaper, how models learn context and complete real workflows, and how AI infrastructure moves toward scalable models like GPUaaS. Lately I have been following what changes when all of that has to leave the screen and act in the physical world.

What I enjoy most is building something new with people and taking on challenges that push me beyond what I already know. I believe the best challenges make us better, and I want to keep learning, growing, and becoming a little better every day.

01. Experience

From research to revenue, role by role

2026 – 2028

Boston, MA

MBA Candidate

Harvard Business School, Class of 2028

Sharpening the management lens on a decade spent commercializing frontier AI

Summer 2026

MakinaRocks, Seoul

Pre-MBA, Physical AI Product & Investment

AI Strategy Division

From the investment case for a drone company to the architecture for defense AI

→ See Project e

2023 – 2026

SK Telecom, Seoul

AI Strategy Manager

Global AI Strategy Division

Led SK Telecom's generative-AI business, from a telco-first LLM to frontier-AI investments and GW-scale data centers

→ See Projects a & b

2020 – 2022

SK Telecom, Seoul

AI Product Manager

Global Strategic Partnership Division

Took the AWS / Alexa computer-vision platform to market, then into Sullivan Finder

→ See Project c

2019 – 2020

SK Telecom, Seoul

Sales & Marketing Strategy Manager

Mobile Business Division

Grew 5G sales 10.9%; Boost Park, a Gen-Z campaign, lifted AI subscriptions 7.8%

Education

Seoul and Tours, France

Bachelor of Business Administration & French Language and Literature

SungKyunKwan University

GPA 4.1 / 4.5 (Top 5%), fully funded exchange at the University of Tours, France

02. Signature work

The full stack, end to end

Four layers of the generative-AI stack, one line and the proof for each. Tap a layer to open the evidence; the full story lives in the projects below.

01

Foundation Model

Built the industry’s first telco-owned LLM, then invested behind it

+

Identified the need for a foundation model the operators could actually own, and built the alliance to fund it. Telecom carries national-scale data under the tightest regulation of any industry, so internal control and data sovereignty ruled out every commercial model on the market.

Brought four global carriers to the table, negotiated terms with the model providers, and stood up a vertical that did not exist before: the industry's first telco-specific LLM, governed by the telcos themselves. Then took the same thesis into capital, working on SK Telecom's strategic investments across the foundation-model and platform layers, from sourcing to deal terms.

Deutsche Telekom ↗

RCR Wireless ↗

SKT Newsroom ↗ Deep dive → Project a

02

Platform

Founded GTAA & Syntelligence: $37.5M JV, $2M revenue in 6 months

+

Productized the telco LLM at the API layer (SaaS / private / on-prem), then structured the five-party joint venture that turned it into a company. Flagship use case: detecting spam calls and deepfake voice fraud.

syntelligence.ai ↗

Four builds across the generative-AI stack, and one where it meets the physical world

a.Foundation & Platform

Building the world's first sovereign telco-LLM alliance

Global Telco AI Alliance (GTAA) → Syntelligence, SK Telecom with four global operators, 1.3B subscribers

Identified the need for a telco-specific model, built the five-operator alliance to fund it, and turned it into a company.

Challenge

Global operators needed generative AI, but no foundation model met their data-sovereignty rules, and none had the scale.

Approach

Secured four founding operators, negotiated terms with the major model providers, then productized the LLM at the API layer.

Outcome

$2M revenue in six months with eight enterprise clients, and $37.5M to establish Syntelligence, the first five-party telecom JV.

JV agreement signed @ DTW24 ↗

#1 in AI, Omdia ↗

Syntelligence ↗

inView post ↗ Global Telco AI Alliance launch

GTAA unveiled at MWC 2024

in ↗

CEO Summit, JV signing

Backed by global leaders

syntelligence.ai ↗ Syntelligence brand

Syntelligence, the JV brand

inView post ↗ GW-scale AI Data Center

APAC's gigawatt-scale AI hub

SKT Newsroom ↗ SKT HAEIN AI Data Center, GPUaaS

HAEIN, SKT AI DC GPUaaS

AIDC expansion → 1 GW ↗

Omdia, Sovereign AI Infra ↗

b.Infrastructure

A gigawatt-scale AI hub for APAC: "APAC's Stargate"

GW-Scale AI Data Center, SK Telecom with NVIDIA, Amazon & Microsoft

Moved SK Telecom beyond colocation to a GPU-as-a-Service hub, then elevated it into a national agenda.

Challenge

AI compute demand in Asia was outstripping supply, but the region had no gigawatt-scale GPU hub.

Approach

Shaped a GPU-as-a-Service model with NVIDIA, Amazon & Microsoft for a gigawatt-scale APAC hub in Korea.

Outcome

Reframed Korea as a candidate hub for APAC AI infrastructure, and supported the letter-of-intent process.

c.AI Service

Productizing a computer-vision platform across four verticals

Computer Vision Platform-as-a-Service (CVPaaS), SK Telecom with AWS & Amazon Alexa

Co-developed and commercialized a public-cloud computer-vision platform with AWS and Amazon Alexa.

Challenge

Video data is heavy, and no one had productized a public-cloud computer-vision platform.

Approach

Co-authored PRDs across four verticals and led the shift from on-prem VMS to hybrid public cloud, then set the go-to-market.

Outcome

A public-cloud platform commercialized across four enterprise verticals, co-built with Amazon.

Business Wire ↗

Mobile World Live ↗

AI Times ↗

inView post ↗ Computer Vision Platform

CVPaaS, co-built with AWS

SKT Newsroom ↗ AWS and SK Telecom develop computer vision platform

AWS & SKT, computer-vision platform

KED Global ↗

Amazon Global Store on 11st

inView post ↗ T Universe subscription platform

T Universe, 1M MAU in 6 months

KED Global ↗

Launch update ↗

d.Commerce & Subscription

Scaling the SK and Amazon partnership across commerce & subscription

Amazon Global Store on 11st & T Universe, SK Telecom with Amazon

Turned the SK Telecom and Amazon partnership into two consumer products and revenue.

Challenge

Three companies with different incentives had to ship products on a case strong enough to win Amazon's commitment.

Approach

Coordinated Amazon, SK Telecom & 11st to launch the Global Store, and built the revenue-sharing model behind T Universe.

Outcome

Global Store launched backed by $30M in committed support from Amazon; T Universe reached one million MAU in six months.

e.Physical AI

Architecting a defense AI product from zero, and pricing a Physical AI deal

Product architecture and investment diligence, MakinaRocks, Summer 2026

Owned the first architecture for the company’s defense AI product, and the technical and financial diligence on a Physical AI acquisition target.

Product

Drew the architecture from the frontline in: a central OS that plans across the fleet, an edge OS that keeps deciding when the link is cut. Benchmarked against Palantir and Anduril.

Investment

Interviewed drone makers and operators across the market to test whether the autonomy stack was real, then modeled it and proposed a valuation below the ask.

Why it matters

A position in Physical AI without buying the risk: staged capital, downside protection, and a rollout that scales only on proof.

Journal → How does AI become an expert in the physical world?

The architecture, in one line

Central OS

Plans the mission, holds fleet-wide context, ships model updates

↓tasking, context, model updates

↑what the field sees and learns

Edge OS

Keeps perceiving, deciding and acting with no connection at all

Diligence, from the field up

Technology DD: autonomy stack, hardware dependencies, demo versus field, read against interviews with drone makers and operators
Financial DD: revenue quality, burn, comparable multiples, milestone-linked valuation

04. Writing & Ideas

Where is the frontier moving next?

AI has largely solved the problem of finding information. This journal follows the gap that is left — turning what a model knows into judgment and action.

Latest02-2

How Does AI Become an Expert in the Physical World?

Take the same three questions into manufacturing, autonomous driving, and defense, and a new dimension appears: prediction.

August 2026, 15 min readEN한국어 Read the essay →

02-1

Okay, Then How Does AI Learn an Industry?

Healthcare, legal, telecom. Very different industries, yet the same three questions keep showing up when we try to build domain-specific AI.

August 2026, 16 min readEN한국어 Read the essay →

Open the journal →

Egocentric AIContextual AIPhysical AIInfrastructureTelco

05. Off the clock

Who I am when I close the laptop

I grew up in Edmonton, Canada, and I was a handful: mischievous, always laughing, always halfway up something. More than half the city was born somewhere else, so difference was never a subject anyone had to teach me.

I’ve always found a lot of joy in being around people, especially when I can make the people I love laugh. There’s something deeply satisfying to me about remembering what someone loves, doing a little something for them, and seeing their face light up.

Julie as a child on stage in a blue costume, arms raised, Edmonton — Edmonton, mid-nineties.A full commitment to the pose.

I build businesses around
frontier AI, where the path has yet to be defined

Zero to one, across borders