Teams at Pluralis Research

Design and implement robust, large-scale distributed ML training systems optimized for low-bandwidth, high-latency environments. Build model-parallel training strategies, checkpointing and recovery, GPU and memory optimizations, P2P networking, NAT traversal, and monitoring to ensure resilient, efficient multi-participant training.

Own end-to-end event logistics (conferences, workshops, offsites, Tech Talks), run CEO travel/scheduling systems, capture post-event learnings in a single source of truth, and provide broad operational support including onboarding, documentation, and establishing a San Francisco office presence.

Work on Agora, a production decentralized training system: design and ship a scoped project, improve concurrent/parallel components, run multi-node training on cloud (AWS/GCP), contribute to production codebase, and present results to engineering and research teams during a 3-month internship.