Standard AI

We use a 3 month cycle called Arcs. At the beginning of the arc we sit as a team and prioritize large scale problems, priorities, and features that we need to get done, then we figure out what we want to get done in the next 3 months. After that we figure out the best way to organize. Sometimes that means multiple small teams, sometimes it means team-wide standups, but whatever it is we tailor our cadence and organization to what will be best for our 3 month goals, rather than finding goals that are best suited for our current organizational structure.

Day to day we use github, pull requests, code reviews, and testing, like most sane dev teams.

We're building a real time deep learning inference engine that has to stream process dozens of camera feeds on a distributed GPU cluster. Keeping this system performant and fault tolerant is a massive challenge.

We're installing a GPU cluster in every store we work with. Monitoring and maintaining these remote systems is a challenge now, but will be a huge challenge as we scale to more stores.

Our backend database has to work in both a fractured state for each store, as well as in a global state, aggregated across all stores. If the internet goes down in a store, our in-store system needs to keep running. Design and scaling this system is an ongoing challenge.

We have to ingest video data of every unique item we encounter in the world. Some stores have 100,000+ unique items. Collecting, maintaining, and serving this data to train machine learning algorithms, especially in a massively distributed setting, is an open challenge.

We have a complex, multi-model machine learning pipeline. When something goes wrong or a model makes an incorrect prediction we need to understand why. Building tools to get insight into this complex chain is a challenging intersection of tooling, UX, and ML.

We're building apps that will be used by shoppers around the world. Perfecting their shopping experience is the heart of everything we're trying to achieve.