Russell Ro

Russell Ro

Production-grade ML and agentic systems. NLP, retrieval, forecasting.

PhD Student @ UC Berkeley

Clinical Trial Eligibility Research

Project Timeline: 2025 – present

Skills: Python, PyTorch, FAISS, sentence-transformers, SQLite, NLP, LLM classification, semantic matching, dual-GPU pipelines

Advisor: Dr. Reza Abbasi-Asl (UCSF / UC Berkeley)

Overview

Evidence-guided clinical trial design and evolution to optimize patient eligibility accrual across 500K+ trials. Includes an LLM benchmark for criteria change prediction and an adoption gap study measuring whether trials implement published broadening recommendations. NLP pipeline with semantic matching, LLM-based classification, and dual-GPU processing.

Key Components

  • Eligibility criteria change tracking: version history analysis across 500K+ trials with 800K+ version snapshots
  • LLM benchmark: evaluation framework for criteria change prediction tasks using retrieval-augmented generation
  • Adoption gap analysis: systematic measurement of whether clinical trials adopt published evidence-based recommendations for broadening their eligibility criteria
  • Semantic matching pipeline: directive-to-trial matching using biomedical embedding models
  • LLM-based classification: adoption assessment at scale using dual-GPU processing