Product Manager — AI Systems

Most AI systems look correct.
They fail the moment behavior matters.

I build systems that behave predictably
in production.

I design systems that make explicit decisions before generating outputs.

Because in production, failure isn't model accuracy — it's what the system chooses to do with it.

Systems don't fail in demos.
They fail in edge cases — and that's where most teams never design.

View Case Studies Get in touch →
Selected impact
97%
System accuracy
up from 88%
<1%
Hallucination rate
via guardrails + RAG
8wk
MVP to production
+ iterative redesign
+8%
Feature usage
targeted segment

These outcomes came from system design decisions — not model improvements.

01

What I focus on

I work on AI systems where correctness alone is not enough.

Most teams optimize for model performance. I design systems that make consistent decisions under uncertainty — systems that still behave correctly when the input is unclear, incomplete, or wrong.

The gap between a model that performs and a system that behaves only becomes visible in production. That gap is where I work.

— When inputs are unclear or incomplete
— When signals are ambiguous
— When the "right" response is not obvious
— When failure modes need to be designed for, not discovered
Point of view

Most AI teams are not building systems.
They are building model wrappers.

That works in demos.
It breaks in production.

Explicit decision logic Controlled behavior under uncertainty Clear handling of ambiguity Failure modes by design System reliability over model accuracy
02

Selected work

Featured Case Study · AI Safety System
Kyrah AI — Designing a Reliable AI Safety System
Model correctness did not equal user trust. Redesigned the system architecture — not the model — to make consistent decisions under uncertainty. Introduced an explicit decision layer between classification and response generation.
97%
accuracy
<1%
hallucination
→
Case Study · Consumer AI
Food Spy AI — Turning AI Output into Usable Insight
Users didn't struggle with getting data — they struggled with understanding what it meant. Conducted 20+ user interviews and translated AI output into structured, readable decision flows that improved clarity and usability.
+8%
feature usage
→
Case Study · Enterprise · NETGEAR
NETGEAR — Designing a Structured Onboarding System
The issue wasn't lack of effort — it was lack of structure. Mapped end-to-end onboarding workflows, defined key states and transitions, and introduced a structured system that improved clarity, ownership, and visibility across teams.
↑
team visibility
→
Case Study · Operations · Arlo
Arlo — Clarifying Internal Workflow Structure Across Teams
Internal workflows existed but lacked consistent structure across functions. Defined workflow requirements, aligned stakeholders across product and operations, and improved how teams coordinated within shared systems.
↑
coordination
→
Clement Lee
03

About

I'm an AI Product Manager focused on turning ambiguous human problems into systems that can make consistent decisions.

My background spans internal platforms and operational systems at NETGEAR and Arlo, where I translated complex workflows into more structured, scalable user experiences.

I now focus on building AI systems that behave reliably in real-world conditions — not just controlled environments.

University of Pennsylvania
MS — in progress
Cornell University
BS
NETGEAR — Internal onboarding workflows and platform design
Arlo — Internal people systems, reporting workflows, and program operations
Kyrah AI — AI safety systems, classification, decision logic, and guardrails