Email, LinkedIn, Resume
About Me
Hello! I’m Kai, a designer and student at UT Austin. I am someone who really enjoys when a design problem, engineering problem, and what people want all turn out to be the same problem.
I’ve worked on problems spanning design systems and documentation for EA Help all the way to Lunar Rover prototypes with NASA L’SPACE. I’m comfortable taking an ambiguous problem, breaking it into small steps that we can work with and communicate to leaders, while still being open for changes that inevitably happen.
I truly love this field and find it so much fun :).
Across this site you can find other artifacts of my work, how I work, and some of my projects.
Enjoy!
Experience:
Service Design Intern, EA
SWE Intern, National Park Service
UX Intern, Merck
SWE Intern, Merck
Service Design Intern, NASA
Education:
Master of Science in Computer Science,
UT Austin, 2025-2026
Bachelor of Arts in Computer Science,
Rutgers University, 2021-2024
Co-curriculars:
NASA L’SPACE NPWEE, 2023
Google’s Software Product Sprint, 2022
NASA L’SPACE MCA, 2022
Weaknesses
If you’re on this site, I’m guessing you’re looking at this portfolio to see if I’m a competent person and if you’d like to work with me, so I should be honest and give you enough information to judge that.
The “what are your strengths and weaknesses” question always sounded like an rehearsal to me. No one ever gives clear answers. When you’re asked about a weakness, people lie and say something quirky like “I’m a perfectionist”, or “I’m not a morning person!”. I personally believe that my answer should include what I think are my honest strengths and my working style, but more importantly, my weaknesses and where I ask for help. I really do value being honest and transparent, and this is the best way I can think of doing so.
My main honest weakness is that with I struggle with gauging progress on many concurrent, abstract things happening at the same time. I don’t miss deadlines or ship bad work, but I often ask my manager or colleagues for sanity checks. In a sense, I very much feel like other people’s sense of time is better than mine. So, the most valuable skill that I’ve learned across all of my jobs that I’ve had is to borrow that function from others. I ask my manager or a friend “am I on track?” often and give them permission to put me on blast if I am not. I share my notes and progress with them explicitly, and just ask for a little bit of time during 1-on-1s and standups. I love beeminder and using commitment devices, and ask my colleagues for more bespoke “stings” when it comes to projects without them spending a lot of time on it. I geniunely like being called out on stuff I’m not on track; I figure out my progress way earlier instead of just before it is due, which is a relief. While I do believe this is my biggest weakness, I have a strategy for it, and address it with enough scaffolding that it’s not an issue for getting things done.
Strengths
Before dumping the entire generic comma seperated list of skills that usually goes here, I think I can generally describe what I can do like this:
I move between research, design, and code depending on what a problem needs. Sometimes that's running interviews and synthesizing them into a framework, sometimes it's building the actual component in Figma or React. I'm most useful when I make sense of an ambiguous problem: turning something unclear into documentation an engineering team can build from, or turning a pile of interview data into a system someone can understand and then act on.
For some other things that I think are strengths or just part of my working style:
- I get genuinely excited about very small, specific things, and it’s why I love this field. Very little things that others don’t find that interesting captivate me, and I love going deep into a field and learning its shape and its edges.
- I (hope I) am emotionally intelligent and can talk to different people from different backgrounds in different contexts. I (again hope I) am thoughtful and really do care about the other people I am around and the thing I am working on.
- I’m really, really transparent and value honesty a lot. I have no problem talking about things I’m struggling with or asking for help from others on something.
- I’m pretty chatty! If I find something interesting, I probably want to tell you about it. I love book clubs and honestly just scheduling 10 minutes one time to learn something new about the other people I work with. I really do care about the other people I work with, and for joining a new place, would like to be a part of a place that gives opportunities for that (no one wants to join and feel like there’s a clique!).
- I default to async communication writing things down and messaging when it's not urgent, because I know how much a random call can wreck someone's concentration, from experience. I'll call when something's actually time-sensitive, or if there’s a dog.
Ok, now you get your skills list :).
List of Skills:
Design: Figma, Zeroheight, design systems, wireframing, responsive design, interaction design, stakeholder management
Research: user interviews, usability testing, co-creation workshops, participant recruitment, Dovetail, data analysis
Technical: HTML/CSS/JS/TS, React, Python, Storybook, design tokens, git, dev handoff, technical documentation, Claude Code/Design
Things you should read
For Designers:
- Mindful Design by Scott Riley
- Just Enough Research by Erika Hall
For Coding:
- The Grug Brained Developer
- SICP by Harold Abelson, Gerald Jay Sussman, and Julie Sussman
- Crafting Interpreters by Robert Nystrom
For Everyone:
- Gödel, Escher, Bach by Douglas Hofstadter
- Small Things Like These by Claire Keegan
- The Dream Machine by M. Mitchell Waldrop
- Stoner by John Williams
- The Grasshopper: Games, Life and Utopia by Bernard Suits
- The Demon-Haunted World by Carl Sagan
Not books but who cares!
- Baba is You - Really, really fun puzzles :)
- Disco Elysium - Amazing story!
- Returnal - Masterclass in environmental storytelling.
Journal Article, 2022
One of the questions we worked with at NASA: how do you design a service for the customer you have today without accidentally boxing out the customer you'll have in ten years? Most human-centered design is great at the first part and pretty bad at the second, you end up building design debt toward a future you never actually mapped.
With the NASA SBIR/STTR team, I helped develop and publish a methodology we called triangulated recursive mapping. It weaves futures research into the HCD process itself, so every near-term decision is quietly pulled in the direction of where things are actually headed, instead of just where they are right now.
It’s a cool parallel to episodic future thinking, how simulating a distant outcome for a distal goal engages prefrontal regions and reduces the urge to discount that outcome and act impulsively.
It feels very much like practicing Mindful Design, which just feels right to do!
It was really fun to write this, and hopefully it’s fun to read too!
Read the paper
pulses, an app and design system
Personal Project, 2025
I made a little design system and a time tracking app for myself :).
For me, input for any of the time tracking tools never really worked because it took more effort to set everything up properly. I wanted something more like iA Writer than Toggl. For this tool, everything is gesture-driven and text-parsed instead, type "writing docs #work @pulses" and tags and projects are pulled out automatically, no dropdowns, no forms. Swipe forward to log the next thing, swipe back to undo.
Most of the actual design work went into a small SwiftUI design system built specifically for this, tokens for spacing, type, and color, with every foreground/background pairing checked for contrast (14.25:1 for primary text, AAA-rated across almost every combination). It's a tiny system, but building it upfront made the rest of the UI work quickly, and the payoff shows up in small details, like debounced ambient tags that float as you type, updating instantly on a space and after a short delay otherwise, so it feels responsive without being jumpy. Really, it was an excuse to make a design system, which I like doing!
Read the writeup
Debiasing Toxicity Detection Models
Research, 2025
Spurious correlations are one of the more persistent but intuitive failure modes in ML: a model learns to rely on a pattern that happens to predict the label in its training data, without that pattern having any real connection to the thing it's supposed to be detecting. The shortcut it found doesn't generalize past the dataset it learned it from.
Toxicity detection is a clean example of this. Models trained on datasets like Civil Comments learn to associate specific words with harm, rather than learning what harm actually is. A model trained this way will score words like "damn" as more toxic than an explicit death threat, because within its training data, profanity co-occurred with toxic labels far more consistently than threats did.
I worked on two interventions to correct for this. The first, composite scoring, changes what the model is optimizing for during training, reweighting the target so explicit profanity contributes less, and severe harm, threats, and identity attacks contribute more. The second, adversarial data augmentation, expands the training set with synthetic examples specifically designed to break the spurious correlations: identity terms used neutrally or supportively, profanity used positively, reclaimed language, and obfuscated threats that the baseline model was missing entirely.
Both interventions reduced identity- and profanity-based false positives substantially, without giving up in-distribution performance on the original dataset.
But overall, I think it’s one of many examples about needing to be conscious when using AI; it’s not just something that shoots out correct answers and code everytime. If we’re going to use it, we need to make sure it doesn’t propogate harm and isn’t just encouraging cognitive surrender.
Read the paper
Digital Garden
Personal Project, Ongoing
I have a digital garden! (see digital gardeningl) I’ve put thousands of notes into it, and am at the moment moving them all into a shiny mind map within Obsidian Canvas. The garden is made with Quartz, written in Obsidian, and hosted with Cloudflare.
At the moment, I have made a little interactive graph of my notes for last year that I think is pretty cool while my notes are bare. Take a look!
Visit the garden
See the graph
uefn-stuff
Sandbox, 2025
I built a procedural city generator in Unreal using Verse, where Conway's Game of Life decides what gets built. Alive cells become buildings, dead cells become parks, and buildings that survive long enough stack upward, so stable clusters grow into skylines while unstable areas keep churning.
Most of the actual work was performance: moving from cell-by-cell updates to batched region updates dropped the SDF recalculation from O(n²) to O(n), which was the difference between smooth and stuttering.
Alongside the generator, I spent some time in Unreal's terrain sculpting tools with a few friends, shaping landscape by hand and layering in atmospheric effects like time-of-day lighting and fog. It was fun!
Read the writeup