Kingston

I am a fullstack product minded engineer. Previously I spent 2 years at TwelveLabs, where I worked on video editing agents for media enterprises built on our video cognition system. On the side I'm building buysignal, where agents help source and sell luxury handbags. Before that I did spatial reasoning research at AI4CE while studying CS at NYU.

I pride myself in my product and design engineering thinking. I got into coding in 2015 by jailbreaking my iPhone 5s, where exploring Cydia apps was my pastime. In high school I swam competitively and posted best times of 23.32 in the 50m free (relay) and 29.63 in the 50m breast. Overall I was ranked 64th in China in 2019.

I'm optimistic about AI. What interests me is where it fits into everyday work — luxury resale still pretty much runs inefficiently across spreadsheets and dms for things like inventory tracking or price tracking. On top of building a good platform, it's crucial to have agents that learn to make better decisions over time (like humans do). In doing so, it was necessary to evaluate the performance of the agents on realized margin. So I built an automated system on the language layer to help continuously update the agent context that made more accurate buyout predictions with multi-armed bandit techniques against real data we collected. Naturally, that led me to take a stab on RL env task creation for domains in finance, long-horizon code, and in CAD modelling, which I now spend the majority of my days playing around with.

我是一个产品向的全栈工程师。之前我在 TwelveLabs 待了两年，负责基于我们的视频认知系统给媒体企业做视频编辑智能体。业余我在做 buysignal.com，用智能体帮忙找货和卖奢侈品手袋。大学时期，我在 NYU 学计算机，同时在 AI4CE 实验室做空间推理研究。

我为自己的产品和设计工程思维感到自豪。2015 年我给 iPhone 5s 越狱开始写代码，在那儿逛 Cydia 上的应用是我当时的消遣。

高中时期，我曾接受竞技游泳训练，50 米自由泳接力分段最好成绩 23.32 秒，50 米蛙泳最好成绩 29.63 秒。总体来说，2019 年我全国排名第 64。

我对 AI 保持乐观。我关注如何把AI融入人们的日常工作。奢侈品转售基本上还是很低效，靠表格和私信做些库存跟踪、价格跟踪之类的事。平台本身要做好，关键是让智能体能随着时间把决定做得更好（像人那样）。要做到这一点，必须按已实现收益率来评估这些智能体。所以我在语言层搭了一套自动系统，用多臂老虎机方法对照我们收集的真实数据，持续更新让买断预测更准的智能体上下文信息。最近我开始尝试做强化学习的内容，主要在金融、长程代码和 CAD 建模这些领域里应用。