CV

Machine Learning Research Collaborator — Cost-Aware Clinical Phenotyping

Dr. Vivek Rudrapatna's Clinical Data Science Group, UCSF

• Independently scoped the cost bottleneck in EHR phenotyping and proposed a confidence-gated cascade as the solution.
• Architected a four-stage pipeline: CPU BioBERT disease-NER → entity-guided retrieval → confidence-gated escalation to a frontier LLM → mention-level precision/recall/F1 evaluation with a token-cost proxy.
• Early NCBI Disease pilot: escalating ~35% of notes raised precision 0.62 → 0.75 while consuming ~24% of the tokens a full frontier-LLM approach would require.
• Engineered CPU-only, no-GPU deployment so the approach runs inside hospital firewalls; obtained PhysioNet credentialing for MIMIC-III/IV.

Data Science Research Assistant, Clinical Research

In collaboration with Stanford-affiliated researchers

• Trichoscopic oculomics: deep-learning classification integrating spatial transcriptomics and metabolic epidemiology to reframe androgenetic alopecia as a systemic screening opportunity.
• Endometriosis severity: supervised ML (random forest, gradient boosting, logistic regression) for preoperative phenotyping from EHR-derived features.
• End-to-end clinical preprocessing — imputation, encoding, SMOTE, cross-validated evaluation — with results presented to clinical stakeholders.

Machine Learning & Data Science Research Assistant

Washington Lab, UCSF

• Systematic review cataloging architectures (CNNs, LSTMs, transformers), datasets, and benchmarks across 50+ studies on home-video ML classification of autism and ADHD.
• Built Python tooling to standardize and visualize sensitivity/specificity/AUC across heterogeneous reporting formats; manuscript in preparation.

Machine Learning Scientist

Optelligence — Berkeley, CA

• Designed and trained supervised regression/classification models (XGBoost, neural networks) for performance prediction, cutting physical testing cycles by 50%.
• Built end-to-end preprocessing pipelines and an A/B testing framework with statistical significance testing, improving deployment efficiency by 25%.
• Applied CNNs and graph neural networks to extract latent structure from high-dimensional experimental measurements; SQL/PostgreSQL extraction and stakeholder dashboards.

R&D Engineer

Amgen & ThermoFisher — Thousand Oaks & Pleasanton, CA

• Automated QC workflows in Python/SQL, maintaining 99.8% assay accuracy across 100+ daily COVID-19 diagnostic samples.
• Built SPC dashboards for HPLC/UPLC instrument drift, enabling early anomaly detection and reducing reagent waste by $50K annually.

Research Assistant — Progesterone Biosynthesis (iGEM)

University of California, Santa Cruz

• Engineered Y. lipolytica for progesterone biosynthesis; designed a 5-gene cassette for heterologous expression.
• Executed Gibson Assembly, yeast-mediated cloning, and Cre-Lox strategies, validated by colony PCR and Sanger sequencing; co-authored a 60-page research report.

Scientific Researcher & Research Consultant

Apex Energetics — Irvine, CA

• Maintained a structured literature database of 200+ papers/month across 15+ nutraceutical compounds with NLP-assisted tagging.
• Synthesized evidence across RCTs, meta-analyses, and in-vitro studies into quantitative reports, saving research leads ~20 hours/week.

MS, Data Science & Analytics

Georgia Institute of Technology — Atlanta, GA

BS, Bioengineering / Bioinformatics

University of California, Santa Cruz