Current internship in fall 2024 bringing AI to a legacy strategy/design consulting company. Writeup soon.

AI strategy consulting

What best practices help tech companies efficiently develop new AI features? A product management internship in the summer of 2024 at Qualtrics in the EX AI team.

AI at Qualtrics

Spent five weeks in early summer 2024 working out of the NFDG office to test popular Silicon Valley foods and beverages for plastic chemicals. We wanted to know: How many plastic chemicals are we consuming, and how bad are they for us?
 
 Learnings about plastic chemical research:
 <ol>
 <li>Plastic chemical researchers are very gracious and open to sharing their learnings, even with naive newcomers. In our initial chats with experts, I was really pleasantly surprised to experience such consistent support and enthusiasm for what we were trying to do. </li>
 <li>There are no 'smoking gun' studies in plastic chemical toxicology for humans (or if there are, we were unable to find any after months of research and speaking to experts!). A causal link is impossible to prove ethically, and even if researchers found a link in a lab setting, real life effects of plastic chemicals in context with co-exposures to other chemicals could look very different. </li>
 <li>Plastic chemicals are ubiquitous. Not only do they exist in all kinds of human-made things like clothes and shampoo, they also get passed into the environment — they're in the air, the water, the land. </li>
 <li>Lots of people can do a 7/10 job at plastic chemical testing, but only a few can do a 10/10 job. Eliminating sources of contamination is the biggest technical challenge in plastic chemical testing — even something like wearing scented hand lotion can throw off results. Someone who does a 10/10 job more efficiently identifies and removes sourced of contamination, using judgment they've built through extensive practical experience. </li>
 <li>In science reporting it's possible to write something that, while objectively true, gives the wrong impression because it lacks the proper context. A paper could have something like an alarmingly high number quantifying phthalate levels, but on closer inspection that number is actually in nanograms and orders of magnitude below the levels of phthalates that were linked with negative effects in the literature.</li>
 </ol>
 
 Other learnings:
 <ol>
 <li>Don't confuse kindess for clarity</li>
 <li>Post weekly summaries of your work</li>
 <li>Create initial assumptions and find a way to test them ASAP</li>
 </ol>

Tested 300+ samples of popular Silicon Valley foods and beverages for plastic chemicals.

PlasticList: testing foods for plastic chemicals

<h3>Prompt: Design something that increases repurposing/reusing in New York City</h3>

 <h3>Why community building? Every New Yorker we interviewed indicated a need for more social connectivity</h3>
 We conducted six exploratory interviews with young adult New Yorkers and got open to any insights they might share, but were surprised when they all indicated a common unmet need before we even mentioned our specific focus on repurposing/reusing.
 <img src="/images/hangout/community-quotes.png" alt="alt text">
 

 Repurposing/reusing doesn't always have to refer to physical objects, so we decided to follow our users and keep digging, even if it wasn't the direction we expected. To further explore the problem space and understand why New Yorkers' social connectivity needs weren't currently being met, we returned to our initial interviewees and conducted ethnographic interviews about how they socialized.
 

 <h3>Before we became more comfortable with each other as a team, AI acted as a social lubricant by providing a common ground</h3>
 While we pursued this design project, we also explored how using AI as a generative teammate affected our team's interpersonal dynamics and ideation. We used ChatGPT to aggregate insights from the ethnographic interview transcripts and prompted it to propose several problem statements using those insights. We found that AI was especially helpful when we were first getting to know each other as a team: its outputs acted as a common ground for us to iterate from.
 <img src="/images/hangout/community-problem-statements.png" alt="alt text">

 After several rounds of ideating solutions, reevaluating the problem space, and returning to users, we refined the problem statement as follows:
 <blockquote>
 Young adults in NYC lack opportunities for spontaneous networking and making new connections due to limited availability of open facilities outside of work hours, absence of built-in communities, and a lack of awareness about available options for socializing outside of their work environment.
 </blockquote>
 

 <h3>AI's training limitations meant it didn't contribute any new or interesting ideas during brainstorming</h3>
 Even though we went through several rounds of prompting ChatGPT, it only proposed solutions we'd already seen in our background research, which makes sense: AI can only ever be as original or creative as its training data. If a problem still exists, it's probably because there aren't enough creative solutions out there. So ChatGPT won't be trained on those sorts of innovative ideas, and it won't generate those sorts of innovative ideas. But how can we train an AI to be more creative than we are ourselves? One approach, which we're already seeing, is to train AI on data generated by other AI, but this kind of model would be very likely to amplify bias/noise in the original data.
 

 <h3>Ultimately, human-led brainstorming was more interesting and thought-provoking</h3>
 We tried to do Crazy 8's, but it ended being more like Crazy 6's... Time pressure forced us to be more creative and outside-the-box, which doesn't happen with AI.
 <img src="/images/hangout/brainstorm.png" alt="alt text">
 

 <h3>Our users felt most compelled by an idea called Hangouts that leverages friends of friends</h3>
 NYC is the densest American city, yet over half of New Yorkers report regularly feeling lonely. Hangout maximizes your opportunities for chance new connections by leveraging your friends of friends. When someone in your Hangout network opens a Hangout at a public space like a cafe or park, everyone in their network gets an invite, including you. You can accept and send the invite to your own network. Now you can go to Hangouts with a mix of old friends and new faces, simplifying your journey to make lasting connections in the busy city of New York.
 <img src="/images/hangout/frames.png" alt="alt text">
 

 By turning the verb 'hang out' into a noun, we make Hangouts the object rather than the other person. When you onboard to Hangout, you add any existing friends you'd like to go to Hangouts with.
 <img src="/images/hangout/open.png" alt="alt text">
 You can get Hangout invites directly from the opener, or through a mutual connection. Here, Pol B started a Hangout and sent it to Harry West, who sent it to Claire.
 

 Play around with the interaction <a href="https://www.figma.com/proto/KJ1WPoduLQOvvlFpdfy5bx/hangout?type=design&amp;node-id=35-5394&amp;t=euczeK312WeqS8wv-1&amp;scaling=scale-down&amp;page-id=0%3A1&amp;starting-point-node-id=35%3A5394&amp;show-proto-sidebar=1&amp;mode=design" target="_blank">here</a> - just press and hold on 'yes' or 'send'.
 

 Once you've attended the same Hangout as someone, they get added to your Hangouts network. Each edge is color-coded by the Hangout you met at, creating a colorful 3D network that grows as you meet more people. The more connections someone helps you make, the bigger their node gets.
 <img src="/images/hangout/friends.png" alt="alt text">
 

 Our friendships and relationships with other people are what make us human. Hangouts helps you make meaningful connections and find community.
 

 Explore the prototype more <a href="https://www.figma.com/proto/KJ1WPoduLQOvvlFpdfy5bx/hangouts?type=design&amp;node-id=35-4773&amp;t=HXDWrU334M2sTjnj-1&amp;scaling=scale-down&amp;page-id=0%3A1&amp;starting-point-node-id=35%3A4773&amp;show-proto-sidebar=1&amp;mode=design" target="_blank">here</a>.

How can we make better use of existing infrastructures, both physical and social, to help New Yorkers build community? And how does the presence of AI as a teammate affect the design process? A project for Harry West's Human-Centered Design and Innovation course.

Hangouts: meet your friends of friends

In 2023 I spent 200 hours as a volunteer emergency medical technician (EMT), where I learned that EMTs are really just PMs/designers/needfinders/interviewers turned to 100. Designers, for example, interview their users to develop a deep understanding of their needs, brainstorm ideas, and create solutions that meet those needs, over weeks or months, in low-pressure settings where mistakes are part of the process. EMTs develop rapport with their patients to understand their symptoms, decide on an intervention, and administer treatment in minutes, in extremely high-pressure settings where mistakes can do serious damage. Here are some practices I learned:

 
 <img src="/images/emt/image.png" alt="alt text">
 

 <h3>Interview effectively by finding and following the story</h3>
 By treating all kinds of patients, from college students to older adults, in high-stakes contexts, I've learned how to tailor my approach to the individual user, ask thoughtful questions, and dig deep until I understand the entire story behind a problem: what we call the 'line of questioning.'

 

 <h3>Advocate for users by predicting needs</h3>
 At the emergency department, EMTs go out of their way to advocate for their patients to hospital staff to make sure their stay is as comfortable as possible, whether that's prioritizing the patient to get a bed sooner or requesting cold packs for an injury. Sometimes I was the only person anticipating my patients' needs and ensuring those needs were met. My guiding question was always, "What can I do to make this experience better for the user?"

 

 <h3>Work where I'll have the most impact by reading my team's needs</h3>
 This goes hand-in-hand with user advocacy — both skills require a high level of situational awareness and constant recalibration based on feedback. When four EMTs are all treating one patient, things can get hectic. I learned when to step back to take care of smaller details, and when to take on higher-level responsibilities if I noticed a gap in our care.

 

 <h3>Build a culture of feedback by making space for high-frequency feedback sessions</h3>
 A lot of patients are unwilling to give feedback to EMTs even if they're in pain, because they don't want to seem 'difficult.' I learned to proactively ask for patient feedback, evangelizing patient interaction from the start as a two-way conversation rather than a one-sided interaction. I also gave and received feedback from the rest of the crew in scrum-like debriefs after each call where we reflected on what went well, what could have gone better, and our goals for next time.

Volunteered 200 hours providing free emergency medical care to members of the Columbia University and Morningside Heights community, including intervention decision, administration of intervention, and transport.

Emergency medical services

Play our recording <a href="https://drive.google.com/file/d/11f7mMdXiLf-m353ZbHOPJuCAmuT4ReKa/view?usp=sharing" target="_blank">here</a> as you read! (Sorry, you'll have to manually open it in a new tab for now because I need to convert my pages from MD to HTML...)

 

 Writeup coming soon!

How many classically-trained violinists does it take to learn jazz improvization? Culture and community-building via music, and why the best teams are like jazz ensembles.

Jazz violin

Because of my contract, I won't show any specific wireframes or artifacts I worked on. Instead, here are some of my takeaways.

 

 As demographic trends continue to shift, we'll need to create better ways to care for an aging global population. AI helps us create more personalized solutions for older adults that fit their unique needs. It also enables us to deliver low-cost services that once needed to be delivered by an expensive expert.

 

 Tomorrow's seniors are not necessarily the same as today's - design for your future customer. Consider large-scale trends to predict their needs 1, 5, and 10 years from now.

 

 It's easy to get swept up in the glamour of a new and exciting technology, but cutting-edge tech is useless in a bad design. When in doubt, always go back to the user to validate decisions.

 

 How do we create solutions that not only address older adults' problems, but holistically increase their well-being? My work this summer was only one answer, and I hope to continue working on more in the future.

Cognitive stimulation therapy (CST) has proved to be more efficient than drug therapies in treating mild/moderate dementia in older adults. How can we use AI to make CST more personalized and accessible? An internship in the summer of 2023 at a stealth agetech startup.

AI agetech startup

Below is the original write-up about the project I had here: 
 Continuing to live on this planet as a species will mean concentrating into ever-denser cities. Urban planners are already designing the sustainable megacities of the future, with all the features we need to thrive except one: nature interaction.

 

 No matter how technological we become, we'll always have within us an innate need for meaningful time in nature.

 

 But urban designers don't even know how to evaluate authentic nature interaction - forget about designing for it.

 

 Along with the HINTS Lab at the University of Washington, I:
 <ul>
 <li>developed a framework for evaluating urban nature design (greenspaces, parks, riversides, etc.)</li>
 <li>collected hundreds of images of urban nature designs from Google Street View</li>
 <li>labeled images with a qualitative score</li>
 <li>and used them to fine-tune a pre-trained computer vision model that quantitatively evaluates urban nature design</li>
 </ul>

 With this tool in hand, urban designers don't need to be experts in nature interaction themselves to design rich nature experiences. They can use AI trained by experts to generate insights for them.

 

 AI is an inflection point. It'll either lead to innovation, or amplification of the status quo. We need to push for intentional AI applications that create better versions of ourselves, our cities, and the world around us.

Helped with research around buil in 11th and 12th grade. How can we build more meaningful forms of interacting with nature?

Quantifying urban nature designs with AI

This site is intended to be a personal archive where I keep track of my writing, both nonfiction and creative. For me, writing = thinking. My nonfiction writing is where I process and internalize my experiences related to work/a future career. It also has the benefit of being something for potential employers to look at, but I don't want that to be the main purpose of this site. My creative writing is a way for me to process and explore my own (notoriously bad) memory. I'm really interested in how recalling things through creative writing filters and changes those experiences.

 

 The people who visit this site are probably going to fall into two different groups: 1) potential employer and 2) someone I know personally. If you're a potential employer and you're reading this, I hope you're liking what you're seeing so far. If you're someone I know personally, why are you stalking me 🤨

Designed in Figma, built with Next.js, and powered by lots of matcha 💚