I'm a Machine Learning Engineer at Checkr, where I fine-tune large language models, build agentic systems, and ship them to production. My work spans LLM fine-tuning and distillation, evals, and multi-agent applications for code review and automation. I hold B.A.s in Statistics and Economics from UC Berkeley and an M.S. in Software Development from Boston University. Based in the Bay Area, I spend my free time hiking with my dog, exploring new board games, playing pickleball, and hosting poker nights with friends.
Fine-tune LLM LoRA adapters for classification and extraction that match or beat frontier models on latency and accuracy, deployed to a shared multi-LoRA GPU serving thousands of requests per hour. Built and shipped a company-wide multi-agent code review service that catches 50% more issues than the baseline, plus a merge-request manager agent that responds to review comments, resolves pipeline failures and rebases, and escalates key decisions to humans.
Engineered a high-performance fuzzy matching algorithm achieving 99.5% accuracy with a 20ms search time, plus an end-to-end statute research agent with a FastAPI server, MCP integration, and web UI, enabling RAG for dynamic, contextual searches. Scraped 2 million state statutes into a Postgres database with vector embeddings for semantic search.
Designed PyTorch experiments to evaluate the relationship between ensemble size and accuracy for ensembles of convolutional image-classification models, and conducted literature reviews on the trade-offs between compute, training time, and performance for neural network ensembles.
Built an ETL data pipeline and a time-series neural network with a transformer architecture to detect anomalies in cloud costs, surfacing 45% more anomalies totaling $13.0M in annualized cost increases. Designed an algorithm to quantify and autogenerate explanations for major cost variances.
