tien · shiang · tsai
Tianxiang Cai 蔡天祥
Nomad in Beijing, New England, Shanghai, now based in Seattle. I’ve gone by “Dave” in English since I was five. Either name works.
👋
The
short
version:
and
zero
Oscars.
A
with
zero
pixels.
A
campaign
as
the
intern
who
ate
the
research.
from
zero
to
3,000
users.
Off
the
clock
I
run
and
shoot
because
I
enjoy
waiting
two
weeks
to
see
my
mistakes.
How I got here, and how I work
I came up through video, and it still shows
A high-school English assignment sent me down a Final Cut Pro rabbit hole, and a decade of video, motion, and VFX followed. At Tesla I co-produced 13 films that went on to reach more than 5 million views.
That decade isn’t a fun fact on my résumé. It’s why I think about products the way an editor thinks about a cut: what does someone need to see before they’ll care?
Last year that instinct fixed something real. Dynery’s onboarding was clean, fast, and quietly starving activation, because we asked people to build a profile before we’d shown them anything worth building it for. The fix wasn’t better craft. It was establishing before asking.
Beijing(2000)
Tesla (2021)
UW MS HCDE (2024)
What I’m like to work with
I designed Dynery’s onboarding in 2025, watched it underperform, and rebuilt it in 2026 against my own decision. Read the onboarding case.
I stay with the work until it ships.
On DyneryPay I owned the monetization logic, the shared-bill permissions, and the privacy calls, not just the screens, and stayed on it through production. I’m most useful when the product decision, the interface details, and delivery stay connected. Read the DyneryPay case.
I go deep too early.
Given a problem with a lot of edge cases, I’ll over-explore before the decision is settled. So I state the decision we need to make first, time-box the exploration, and lead reviews with the recommendation instead of the reasoning.
How I build
judgment is the job
My title says designer. In practice my PM routes production front-end through me: I use Claude Code to scale Dynery’s 350+ component design system and to produce React Native, then take it through my own QA before it reaches engineering.
Generation is the easy part. The job is catching what it gets wrong, so I encoded a provenance, token, state, and accessibility audit into the agent itself, gated before handoff.
Design for humans
not the trend cycle
We’re three years into the AI gold rush, and most products are bolting a language model onto something and calling it innovation. I work on an AI product. The question I come back to every time is whether the AI is solving a problem here, or just performing innovation.
I’d rather ship a slower, considered feature that earns trust than one that wears this season’s vocabulary.
Beyond design
You can find me…
I studied Digital Media & Design at UConn. Some of that work, on mental health and freedom of expression, was recognized at international human rights forums.
These days: half marathons (1:47 PR), DJing electronic music, film photography, and trail running the Pacific Northwest.
Made in Seattle with ❤️