Technical Details
Our backend tech stack is built on AWS, managed with terraform. We deploy
several node (typescript) and python binaries through ECR and K8s. The
binaries communicate with each other using SQS, and store data to a S3 blob
store and DocDb data store. We depend on ML tooling (primarily Torch on our
own infra, but potentially serverless inference service integration with
langchain as well as ONNX/TorchLib backend), advanced usage of OpenSearch. We
utilize an event-sourcing model to support multiplayer and sharing.
You don't need to be familiar with ALL of these technologies, but you'll
probably struggle without some experience with at least a few of these.
The SOOT backend is responsible for gathering heterogeneous data from various
sources and integrations and processing tens of thousands of unique pieces of
data quickly, while maintaining high availability to support a seamless
frontend user experience. This means an intense focus on speed, integrity, and
scale, with an eye towards maximizing parallelism, such that the system can
easily consume resources as they become available. The backend is also
constantly growing in terms of feature complexity – we are rapidly bringing
new features online, all of which support step function improvements to SOOT.
Much of the role is about developing coherent abstractions and thinking about
data lifecycles. The SOOT backend is responsible for managing a lot of widely
varying behaviors, from ML processing to data compression to basic CRUD. To
support all of this while ALSO growing our feature list, we heavily focus on
reducing complexity.