Andy Konwinski

Konwinski earned a computer science PhD at the University of California, Berkeley. During graduate research, he contributed to Hadoop, co-created Mesos and worked on Google’s Omega cluster scheduler. At Databricks, he served as VP Product AI/ML and co-authored Learning Spark. He was a founding president of Perplexity. He now advises and invests in technical founders through Laude Ventures and supports computer-science research through Laude Institute.[1][2]

Sharing a cluster without one universal scheduler

Mesos lets different computing frameworks share a cluster while keeping their own scheduling decisions. The cluster manager offers resources to each framework, and each framework decides which offers to accept and which tasks to run. This two-level design separates allocating resources from choosing application-specific work. Konwinski co-authored the paper, whose experiments examined utilization, data locality, scaling and failure resilience. The work helped make shared infrastructure practical for diverse distributed applications.[3]

Measuring what agents can do

At Laude, Konwinski helped create Terminal-Bench and Harbor with Ludwig Schmidt’s Stanford group. Terminal-Bench evaluates agents on concrete tasks, while Harbor provides a framework for specifying sandboxed tasks and running evaluations or optimization. Keeping the task environment explicit helps make a result reproducible and makes it possible to compare a model and its surrounding agent software on the same work.[1][4][5]

Konwinski co-teaches Berkeley’s Research to Startups PhD seminar. His work at Laude combines support for ambitious research with investment in technical founders who can turn new ideas into products. His own career follows that route from distributed-systems research into companies and then into new research infrastructure.[2]