CV
Education
University of California, Berkeley — M.S. Information and Data Science (MIDS), GPA: 4.0 Expected August 2026. Relevant coursework: Applied Machine Learning, Natural Language Processing, Computer Vision.
University of California, Riverside — B.S. Computer Science, GPA: 4.0 June 2025.
Research Experience
Graduate Research Assistant — UC Berkeley (Advisor: Prof. Tanya Roosta) January 2026 – Present
- Co-authored systems-level survey on hallucination in medical LLMs, synthesizing 50+ papers on structural mechanisms, detection, mitigation, and benchmarking through a risk management lens. Published MetaArXiv March 2026.
- Developing novel detection and mitigation methods for LLMs building on survey findings, targeting submission to a top-tier ML venue.
Graduate Research Assistant — UC Berkeley (Advisor: Prof. Cornelia Paulik) August 2025 – Present
- Built RAG pipeline for pediatric medical QA combining dense encoders and sparse retrieval (BM25) with cross-encoder reranking, improving accuracy by 34% and reducing hallucinations by 42% over baseline.
- Developed PyTorch classifiers for age-group segmentation (0–2, 3–5, 6–12, 13–18), achieving 94% accuracy for targeted QA.
Machine Learning Researcher — DASION (Research collaboration with Prof. Weiqing Gu, Harvey Mudd College) September 2021 – Present, Claremont, CA (Remote)
- Developed multimodal depression detection using Wav2Vec 2.0 audio and MT5/RoBERTa text transformers with CTC temporal alignment and multi-instance learning, achieving F1=0.94 on DAIC-WOZ.
- Secured $2.5M in NSF Phase I and II funding for clinical ML research. Led team of 8 engineers building production ML infrastructure on AWS, achieving 99.9% uptime and deploying to 5+ institutions.
- Built interpretable CNN models using PyTorch and SHAP for clinical imaging achieving 93% diagnostic accuracy. Developed BERT-based NLP for clinical documentation reducing manual processing by 60%.
Machine Learning Research Engineer (Contract) — Angel Technologies September 2023 – February 2024, Brea, CA
- Built transformer-based NLP system for automated regulatory compliance detection using fine-tuned BERT models, reducing processing time by 40% and error rates by 85%.
- Developed predictive models using time-series forecasting for customer behavior analysis, improving conversion by 30%.
Publications
Rethinking Medical LLM Hallucinations: A System-Level Survey Matthews, Vankadaru, Roosta, Passban. MetaArXiv, March 2026. Survey arguing hallucination in medical LLMs is a structural property of probabilistic generation. Synthesizes detection, mitigation, and benchmark literature through a systems and risk management lens.
Multimodal Multi-Instance Learning for Depression Detection (Target: NeurIPS 2026) First multimodal MIL framework for depression detection combining MT5/RoBERTa text with Wav2Vec 2.0 audio via CTC temporal alignment. Achieves F1>0.90 on DAIC-WOZ, surpassing text-only MIL baseline (F1=0.88). Directly addresses interviewer bias via strict prompt exclusion.
PedRAG: Retrieval-Augmented Generation for Pediatric Medical QA (Target: ICML Poster 2026) RAG framework combining dense and sparse retrieval with age-specific classification, achieving 34% accuracy improvement and 42% hallucination reduction over baselines.
Professional Experience
Chief Technology Officer — AGMNT February 2024 – Present, San Ramon, CA (Hybrid, Part-time)
- Built TensorFlow recommendation engine and full-stack platform (Node.js, React, MongoDB) serving 10K+ users and 15+ brands, improving conversion rates by 28%.
- Implemented A/B testing framework processing 1M+ daily events to drive product decisions.
Machine Learning Intern — Ambassadore Healthcare Inc. May 2023 – August 2023, Artesia, CA
- Engineered distributed data pipeline using Kafka and MongoDB processing 100GB+ daily clinical data with statistical modeling in Scala/Spark, achieving 75% storage optimization and sub-100ms query latency.
Software Engineer Intern — Royal Majesty Home Care June 2022 – August 2022, Long Beach, CA
- Built AI scheduling system using decision trees and linear programming, reducing overtime costs by $200K annually.
Technical Skills
Languages: Python, R, C++, SQL, JavaScript
ML/AI: PyTorch, TensorFlow, Hugging Face, SHAP, MLflow, Pandas, NumPy, Scikit-learn, AWS (SageMaker, Lambda, S3, EC2), Docker, PostgreSQL, MongoDB
Certifications: AWS Certified Machine Learning — Specialty (2025), Linear Algebra with Applications in ML (Harvey Mudd College)
Grants and Funding
NSF Phase I and II Grants — $2.5M for clinical ML research (2021–2026), in collaboration with Prof. Weiqing Gu at Harvey Mudd College.