Building systems to produce data to train the worlds leading LLMs

Architect and core contributor for Anyscale Endpoints. Working on scalable, performant LLM serving.

Working full-stack with to deliver the best developer experience for Rockset customers.

Improved offline accuracy of offsite conversions mode by 0.5% in offline testing by working to deploy dot selection model. Demonstrated an online revenue gain of 0.3% in A/B testing. This model serves millions of dollars worth of traffic per day.

I worked to improve the reliability of Hive testing for 3 months.

Cut main Hive testsuite runtime by 85%, saving thousands of CPU hours per cluster per day across dozens of clusters

Wrote 18,000 lines of production code in 12 weeks

Built a service to algorithmically add production hive queries to testing suites, improving coverage by 30%