San Francisco, California, United States
2024 — 2025
Seattle, Washington, United States
Led cross-language document retrieval initiative by integrating with a multi-lingual embedding model, enabling search capabilities across multiple languages and expanding platform accessibility for global users.
Developed an intelligent query translation system using Large Language Models to automatically convert natural language search queries into optimized Lucene syntax with semantic re-ranking, enhancing search accuracy. Optimized
prompt effectiveness via chain-of-thought reasoning and iterative testing.
Participated in rotational shifts for Dataverse Search reliability. Triaged production alerts for quality and latency regressions, and coordinated mitigation responses. Identified blind spots in alert coverage and implemented new alerts to increase coverage.
San Francisco Bay Area
Joined after Series A when the company had 15 engineers and 30 employees overall. Built out features across the stack before specializing in a backend/infrastructure role. Tools used include Ruby, Sinatra, PostgreSQL, AWS, and Terraform.
Feature development:
Led project to overhaul filtering, sorting, and text search backend logic for in-product spreadsheet-like view, keeping endpoint latencies constant as database scaled 10x.
Collaborated on implementation of First Event, Last Event, and Next Event spreadsheet columns using synced calendar data, enhancing data accessibility for users.
Collaborated on creation and adoption of in-house logging system, allowing developers to more easily search and read logs.
Database Management:
Optimized SQL query performance and Ruby background job efficiencies through PostgreSQL statistics, restructuring queries, and modifying indexes. Authored popular blog post on using PostgreSQL extended statistics.
System architecture:
Led creation of acontrol plane service to manage a cellular multi-region architecture to better support European companies. Used AWS Step Functions and Lambdas to coordinate data between separate databases.
Led project to populate a dedicated performance-testing environment with fake data and built tools and workflows for simulating production traffic, allowing developers to test features without risking product reliability.
2016 — 2016
San Francisco Bay Area
Improved latency of video replication system and implemented metrics to measure performance increase. Tools used include Golang and Mode Analytics.
2015 — 2015
In Natural Language Processing research team, created workflow for streamlined data collection from human raters in Natural Language Processing research team. Established baseline performance measurements on ATIS dataset using context-free grammars. Expanded feature extraction capabilities of internal Machine Learning tools. Used Python, Google internal tools.
Education
University of California, Berkeley
Master's degree
UC Berkeley