Phoenix is a networked distributed key-value store built using Go and gRPC that provides linearizable semantics. Phoenix is built from the ground up, implementing the Raft consensus protocol (leader election, logging, snapshots, log compaction) within replication groups, duplicate request detection, and shard coordination, and shard reconfiguration.

PHOSS is a platform for conducting fast and robust HPO scheduler experiments using synthetic landscapes. PHOSS can simulate running schedulers on real-life data using a smaller number of samples to create a large amount of synthetic data. At its core, PHOSS is a computationally efficient and expressive exponential decay parametric modeling approach to accurately simulate loss landscapes from real-world datasets for fast HPO scheduler experimentation.

RayICV is an internal cross-validation solution for the Ray open-source unified compute framework library that supports hyperparameter tuning. Specifically, it extends the Ray Train framework by implementing built-in cross-validation in order to emulate the scheduling of jobs based on folds trained and model configuration.

CoffeeOS is a multicore x86-64 operating system that supports kernel task suspension, system calls, virtual memory, and multithreading. Other features include wait queues, caching, prefetching, synchronization primitives such as spinlocks and futexes, and an on-disk file system supporting file extensions and directory trees.

ShuttleDB is a column-oriented NoSQL database that can efficiently handle 10M+ tuples per column with support for selecting, fetching, joining, loading, column arithmetic, and multiple aggregation types. ShuttleDB has built-in modules such as the parser, query optimizer, and catalog, as well as advanced optimizations to minimize IO costs such as primary and secondary indexing with B-trees, custom TCP streaming protocols for large payloads, zone maps, hash joins, and multithreaded shared scans.

YouTube Party is a full-stack web application that allows any number of users to stream any YouTube video synchronously using web sockets and a YouTube iFrame instance. YouTube Party also includes an integrated group chat.

Spotify Curator is a full-stack web application leveraging machine learning to help users discover new songs they love. Spotify Curator calls the Spotify API to securely authenticate users and store their liked songs' attributes as training data. It then performs k-means clustering to determine the probability that the user will like any new song or playlist. Model training and predictions are stored in a distributed task queue and picked up asynchronously by Celery workers. Redis serves as both a message broker and result store for the queue.

RoommateHub is an iOS application providing a centralized platform for roommates to easily communicate and coordinate tasks with each other. Features include secure authentication, shared task lists, interactive roommate profiles, anonymized messaging, and embedded iMessage support.