NAME

Determined - Open-source deep learning training platform for GPU clusters, acquired by HPE

SYNOPSIS

2019 - 2024; Go, Python, PostgreSQL, Kubernetes, gRPC, TypeScript
github

DESCRIPTION

Open-source platform for distributed training, hyperparameter search, and GPU cluster resource management, with a Go master, a Python training harness, and Kubernetes and Slurm resource managers. Joined Determined AI as engineer #14 in 2019 and stayed through the 2021 HPE acquisition, working across the job scheduler, metrics storage, Kubernetes support, APIs, and CLI. Landed 695 merged PRs in the public repo, which has 3k+ GitHub stars.

HIGHLIGHTS

• Built the job queue: persisted jobs, queue position in the WebUI and CLI, then reordering, priority, and weight controls
• Fixed priority scheduler preemption and unified ordering logic across the Kubernetes and priority schedulers
• Moved metrics to partitioned PostgreSQL tables with generic custom-type metrics, summary metrics, and Python SDK support
• Led the PyTorch Lightning adapter integration, with mixed precision support, docs, and examples
• Cut CLI startup time by 80% through dependency tree analysis and lazy imports

MEDIA

SEE ALSO

* projects(7), all projects
* ari(7), the resume
* PDF Resume: Download Resume