let me plan themed dinners with your next city
Hey 222 team — I'm Marvin Tai. I want to be your first community & product intern.
I found you by accident. I was in the App Store looking for something that let strangers meet over dinner, because I'd started building it myself. It turned out it already existed and was better than what I had. So I signed up instead.
I love talking to strangers. Freshman year I spent two months across Senegal, knocking on doors and eating meals with people I'd never met. I've since run a philosophy book club in Charlotte for people from 7+ countries and interviewed founders across the US about why they're really doing what they're doing.
The dinner I still think about started with card games and ended in a two-hour argument about theology. What made it happen was rapport — when people in the room started paying attention to when the table was ready to go deeper. That's the part software can't do yet, and it's the part I'm good at.
I can also build and ship what I propose. I'm not a CS major; I taught myself and shipped two apps solo. Beloved is a devotional iOS app I built, localized, and marketed myself, and I published research at CHI 2025 on legal AI.
I also write. My Substack is about meaning, ambition, and what a life is for.
Why 222: after looking at everyone doing IRL, you're the only one curating the whole night — the group, the venue, the thing people actually do. But going through the funnel myself, I think you're matching on the wrong layer. The form knows my taste well. It asks about what I believe, but not nearly as much — and in my experience shared values carry a table further than shared interests. That's the first of the proposals below. I want to spend my twenties on something I can stand behind, and this is the thing I was already trying to build.
what i'm proposing
Every city has its own social heartbeat. We don't want to just copy-paste our playbook. Brylan Donaldson, head of operations — on the Phoenix expansion
That's the whole job. A playbook scales; a heartbeat has to be found, city by city, by someone who shows up and talks to people. That's the work I've been doing for years. I know how to fill a room, and more importantly how to keep it from going flat at minute forty.
I'd host these every week, each one different in nature — nature walks, philosophy, arts, book clubs. A premise, not just a table. Each theme appeals to a different kind of lost human touch, and it's what gives 222 a texture Timeleft and NomadTable don't have.
Every guest brings one friend. Easier to say yes to, and every dinner brings in new people.
I'm at Duke, which puts me inside the Raleigh–Durham–Charlotte triangle: a market full of recent transplants who don't know anyone, and one nobody is serving. If it's anywhere on the roadmap, I can start finding the heartbeat now.
Three proposals, all from going through the funnel myself — about the distance between what 222 promises and what the product currently feels like.
The curation form already asks about worldview — I'd go further. In my experience running dinners, shared values carry a table further than shared interests: two people who disagree about everything except what matters will talk for three hours, while two people with the same playlist run out at forty minutes. A longer, more layered values section would give the algorithm the layer that actually predicts a good table.
This is a message I got from the 222 iMessage app. Small thing, but I would propose making the texts sound friendlier and more human. Imagine a friend texting you.
222 sells chance, but the current algorithm is optimizing for compatibility. I propose an opt-in toggle that still uses your form and your values, but deliberately builds a varied table — including people who don't share your tastes.
four people who share your taste in books, your neighbourhood, and roughly your week. a good night, and a predictable one.
I'd run it as a capped experiment. The obvious risk is that opting out of matching produces worse dinners, or attracts a self-selecting group that makes them worse for everyone else. So the idea I have in mind: limited cohort, defined window, measured on post-dinner ratings and repeat-booking rate against control. If it underperforms, it dies. If it works, it's a new positioning no competitor can copy without abandoning their own matching story.
things i've made
ready to talk? 👀 mtaibuilds@gmail.com · linkedin