By Brent Goldman - November 18, 2020
I’m thrilled to announce the newest member of the Sisu Engineering team, Mark Yen, as our new Engineering Manager for the Platform team!
Mark joins Sisu from Uber’s Advanced Technologies Group, where he worked on self-driving technology, leading the teams building self-driving dispatch, routing, and fleet infrastructure. Before self-driving, Mark was an early engineer in Uber’s core business and led several efforts around pricing algorithms and its real-time dispatch engine. Prior to that, Mark worked as an engineer on the data team at GETCO, a high-frequency trading firm.
I was lucky to work closely with Mark on and off at Uber for most of the past six years, where we built and led many fun, growth-oriented, high-performing teams together. I’m psyched to partner with him again here at Sisu.
To properly introduce Mark, I asked him to share his thoughts on joining the team, how he tackles challenges, and, most importantly, his hot takes on the bedrock of any startup: pizza.
For me, the biggest draw was the massive opportunity in front of Sisu. For most of history, gathering data was the biggest obstacle to making data-driven decisions—the more you could get your hands on, the better. But in recent years, a profound inversion has occurred, and huge amounts of data are now commonplace, but the tools we have to work with these datasets were built for a smaller-data world. In this new world, the challenge is to quickly find the nuggets that matter out of the avalanche of data you have access to. Sisu is tackling this exact problem, and in the process unlocking data-driven decision making in today’s data-rich world for organizations of all kinds.
You and I saw this problem firsthand on Uber’s Marketplace team. We identified dozens of important metrics, but it was impossible to keep tabs on all of them. Whenever one metric changed while others stayed flat, there was a frenzy as everyone scrambled to find an explanation. At Sisu, I’m excited to have the opportunity to solve this problem for everyone, once and for all.
When I first joined Uber, I was on the Dispatch team, which was responsible for the core trip flow. There, I learned how to maintain scalability and reliability in the face of the company’s hypergrowth.
One problem I worked on was figuring out how we could index driver locations efficiently to ensure that riders were getting matched with the nearest driver when they requested a ride. This was particularly challenging because unlike most geospatial search use cases, in which the things you search for stay put (such as restaurants on Yelp), Uber’s drivers are constantly moving throughout the city. The nearest driver to a rider in one minute very likely won’t be the nearest driver the next. Off-the-shelf solutions weren’t up to the task of dealing with our constant stream of updates, so we had to implement our own. We held driver locations in memory, sharded by S2 cell. Each worker node indexed the drivers within its S2 cells inside a variant of a k-d tree for fast nearest neighbor lookups.
More recently, on the Self-Driving Logistics Platform team I led, we worked on the systems around the self-driving car that allowed it to integrate with the rest of the Uber network. One challenge my team worked on revolved around the “operational domain” that limits how self-driving cars can operate on the road. A self-driving car’s operational domain may dictate that it can only go on certain streets or cannot operate under certain weather conditions. The operational domain can differ from car to car, depending on the type of car or the software version installed on the car. My team built a system to manage all the assignments being performed across a self-driving fleet. Part of its responsibility was to ensure that all incoming rider requests, mapping, or testing tasks were routed to the most appropriate vehicle, accounting for its operational domain and the route it would take.
Sisu has worked hard to get our product to where it is now, crunching through vast amounts of data to deliver faster insights to analysts.
The next phase of our company’s history will focus on making our analysis accessible and valuable to less data-savvy users. We want to deliver more flexible analyses and provide more digestible summaries and visualizations. Getting there will require more computation and more data throughput on all of our backend systems. Sisu’s platform is already pushing the envelope on what a world-class high-performance system can do with big data, and I’m excited to help the team scale our systems even further to take on these new challenges.
I’ve been riding my bike ever since I was a little kid, cruising around my cul-de-sac on my training wheels. But I’ve done very little distance cycling—usually, the furthest I’ll go is across the Golden Gate Bridge to Tiburon. When I’ve tried going farther in the past, I’ve had to deal with cramps and saddle soreness.
But during the pandemic, I’ve rediscovered cycling as a great way to stay active. My goal is to one day complete a century, which is a single ride of 100 miles. A typical century can take 6.5 to 7 hours, quite a bit longer than running a typical marathon. But I’m hopeful that once the pandemic is behind us, we’ll start to see organized rides like this one return.
I’ve got a two-parter: 1) stuffed crust pizza is underrated, and 2) the United States needs to step up its game on pizza crust innovation.
Stuffed crust pizza was invented by Pizza Hut in 1995 in the United States. But since then, not much has