Hongbee Park

About

The short version

I'm a product builder and founder with 8+ years of experience shipping systems used by tens of millions of people across fintech, government, commerce, and AI. My work has always sat between complex systems and real-world adoption: finding the research, translating it into product decisions, and shipping things people actually use.

At Harvard MDE and Scale AI, that focus led me to model behavior. I became interested in a gap that benchmarks often miss: AI can be fluent, safe, and factually correct, yet still fail in the social contexts where people actually use it: workplaces, healthcare, public services, education, and other high-context domains.

I'm now building tools for understanding and improving how AI behaves in real-world human contexts, combining model behavior evaluation, human-centered research, and product systems.

The next AI product bottleneck is not just capability. It is behavior in context.

52M

People reached through national-scale digital products

8+ yrs

Product leadership across fintech, government, commerce, and AI

45+

Frontier models evaluated across reasoning, multimodal, and behavior tasks

200+

Evaluations of model behavior against human expectations

Experience

Where I built things

2025

Scale AI

Researcher, Humanity's Last Exam

Analyzed behavioral patterns across 45+ frontier AI systems. Built structured taxonomy for evaluation and prioritization. Classified failure modes by type, complexity, and trigger conditions. A/B tested rubric formats for SFT annotation quality. Experienced firsthand the gap between benchmarks and real user expectations, which became the motivation for Kairos.

AI Research Model Evaluation

2024 - 2026

Harvard University

Master in Design Engineering

Built 4 ventures across hardware, biotech, and AI. Teaching Fellow for Product Management and HCI courses. Campus Ambassador for Anthropic Claude: organized 5 non-technical hackathons, industry-academy exchange sessions, and published guides on using Claude with non-technical students. Firsthand experience of how real students interact with and feel about AI. XR Club President (60 members, 600+ participant conference with Meta, Nvidia, Samsung).

Design Engineering Teaching Fellow Claude Ambassador

2020 - 2024

Toss (Viva Republica)

Lead Product Owner, promoted to Domain Leader

Korea's leading fintech platform. Joined at 300 employees, grew to 4,000. Promoted to youngest Domain Leader. Led vision, strategy, and roadmap across 5 product verticals and 50+ cross-functional teams. Drove 15M+ user inflow, improved MoM retention from 70% to 85% through behavioral cohort segmentation and rapid A/B experimentation. Digital Platform Government kickoff member: led digital government services for 52M citizens including tax payments, vaccine certification, and public document issuance. Ministerial Commendation under the Presidential Digital Platform Government initiative. Opened new revenue category through investment ad platform in regulatory sandbox.

Product Management Growth Government

2017 - 2020

LG Fashion (LF Corp)

E-Commerce Product Manager

Full-stack PM across 7-domain e-commerce platform: search/SEO, product discovery, checkout, payments, returns, customer service. 2-year call center embed: turned direct user pain into 130+ product fixes, reduced inquiry rate from 15% to 5%, generating ~$770K/month in savings. Launched Korea's first ML-based Virtual Fitting service end-to-end, coordinating with Berlin AI research team. Showcased at CES 2019.

Product Management ML/AI E-Commerce

Approach

How I work

"I start where the user is, not where the system is."

At LG Fashion, I sat in the call center for two years to see what users actually go through. That became 130 product fixes. At Toss, user interviews and behavioral data drove every growth experiment. At Scale AI, I looked past benchmarks to ask what real users experience. Kairos started the same way: not from model architecture, but from what 50+ users across three cultures actually expect.

LG Fashion call center Toss UT/IDI Scale AI Kairos 50+ users

"I move to where I don't belong."

Fashion e-commerce to fintech to government to AI. Each time, a domain I'd never worked in. At Toss, I deliberately rotated through public services, user growth, and investment, choosing unfamiliar territory each time. The pattern isn't restlessness. It's how I learn fastest and find what others miss.

4 domains 4 Toss silos 4 Harvard ventures

"I build to learn."

Four ventures at Harvard in two years across hardware, biotech, and AI. Thailand, India, Dubai, Germany. Not to build four companies, but to repeat 0-to-1 in different contexts until the pattern became clear: every domain has a version of the same question. Is the technology actually helping the person on the other side?

AIRQUA Thailand ChoLab India Medly US Kairos cross-cultural

Want to know more? →

Recognition

Awards and honors