I study how humans and machines reason, and how biology can teach AI to reason better. My work sits where computational neuroscience, foundation models, and healthcare meet: not benchmarking for its own sake, but building systems that solve real problems in places where accuracy, safety, and trust matter.
Trained through a Ph.D. in computational neuroscience and an M.S. in clinical-trial analytics, I move from research question to deployed system: peer-reviewed publications, open datasets, 80+ open-weight model releases, and production AI systems serving healthcare at national scale. I am a Senior AI Research Scientist at UnitedHealth Group, Co-Chair of the UHG Enterprise IRB, and an Assistant Professor-in-Residence in Information Systems at UNLV's Lee Business School.
The through-line is translation: turning hard questions about reasoning, safety, and clinical decision-making into artifacts people actually use. I research, I build, I ship, and I close the loop between what we learn and what the world needs next.
AT A GLANCE
Coding as a weapon · 1,554-prompt malicious-code benchmark; USENIX & NeurIPS targets
CardioEmbed · a leading clinical-cardiology embedding model, targeting the costliest area of U.S. care
Coding as a weapon. A four-paper program separating executable malicious-code generation from defensive security knowledge: a 1,554-prompt consensus benchmark (Fleiss' κ = 0.876), a 10-model behavioral benchmark documenting 36 to 72 pp refusal gaps, a mechanistic abliteration study of separable refusal directions, and a thirteen-corpus systematic review.
Adversarial robustness. TEMPEST replication tested 10 frontier models from 8 vendors with 1,000 harmful behaviors and 97,000+ API queries (96 to 100% attack success for 6 models; extended reasoning cut ASR from 97% to 42%), with companion work on guardrail robustness and instruction adherence across 256 LLMs.
Research acceleration. An LLM ensemble extracts structured protocol fields from ClinicalTrials.gov, doubling retrieval to 83 eligible aging-related tDCS trials with near-perfect agreement (κ≈0.94; ICC≈0.95 to 0.96); a multi-model comparison across OpenAI, Anthropic, Google, and Meta benchmarks clinical-trial data extraction end to end.
Domain models. CardioEmbed and LoRA-adapted embeddings deliver specialized clinical-cardiology text representations.
Safe deployment. Vision-token masking in medical OCR plateaus at 42.9% PHI reduction, while a hybrid vision+NLP defense reaches 88.6%, showing why "masking-only" defenses fail.
Early-warning models. Deep embedded clustering over 1.7B+ open-banking transaction rows detects a roughly 3-month window of gambling acceleration before crisis, flagging 243% higher losses and ~17% more income allocated to gambling, for a projected ~12× ROI on earlier intervention ("Catching the Wave Before It Crashes").
LLM safety. Co-authored peer-reviewed work with gambling-treatment professionals evaluating LLM responses to problem-gambling questions, ensuring support tools reduce harm rather than inadvertently encouraging risky behavior.
deep embedded clustering · open-banking data · LLM safety
doi:10.1007/ s10899-025-10430-x
UNLV IGI · with K. Ghaharian, B. Poston
03
Research Program & Work in Progress
Adversarial evaluation of frontier models
Replicable, scaled measurement of how alignment fails: TEMPEST replication (97K+ queries), guardrail generalization, and instruction adherence across 256 LLMs · released as open datasets.
How models reason · CoT faithfulness
A three-paper line on whether thinking-token traces reflect a model's computation, pitfalls in faithfulness measurement, and what thinking/answer divergence reveals about open-weight reasoning systems.
Code safety & cybersecurity
A four-paper program separating malicious-code generation from defensive knowledge: a 1,554-prompt benchmark (κ = 0.876); a 10-model, 7-vendor benchmark (36–72 pp gaps); a mechanistic abliteration study; a review of thirteen corpora. with Dr. G. D. Moody, UNLV.
Clinical AI at deployment scale
PHI leakage in medical OCR, fairness audits of LLM-based ED triage (EQUITRIAGE), and cardiology embeddings (CardioEmbed, with Dr. Alice M. Matthews), plus scientific oversight across UHG, Optum, and affiliates as Enterprise IRB Co-Chair.
IN PROGRESS · TARGET VENUES 2026
›Is Code Safety Fundamentally Weaker Than Content Safety? Multi-vendor benchmark across 10 coding LLMs (36–72 pp gaps, p < 10⁻⁴²). Target: USENIX Security 2026.
›Is Code Safety a Separate Direction? Mechanistic test of abliteration on coding LLMs with HumanEval/MBPP capability preservation. Target: NeurIPS 2026.
›Two books in progress · The NeuroAI Handbook (100+ papers) and Healthcare Analytics and AI (primary text, UNLV Lee, Fall 2026).
›Reasoning-model safety · sandbagging detection, hallucination probes, machine unlearning, MCP-protocol safety, steganographic reasoning; plus a six-manuscript tDCS / muscle-fatigue line with the Poston lab.
PART II
Publications & Scholarship
04
Publications & Journals
2026
Young, R. J., Fonseca, J., & Poston, B. "From API to Action: A Multi-Model Comparison of OpenAI, Anthropic, Google, and Meta LLMs for Clinical Trial Data Extraction."
Young, R. J., & Moody, G. D. "Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023–2025)."
Gada, E., Pangburn, P., Sahr, C., Jarvis, M. S., & Young, R. J. "Clinical and Cost Outcomes in Medicare Advantage Following Post-Acute Care."
Home Health Care Mgmt & Practice
Young, R. J., Ghaharian, K., Golab, L., Kraus, S. W., Wells, S., & Soligo, M. "Comparing LLM and Human Expert Responses to Problem Gambling Questions."
Springer · AIR-RES'25
978-3-032-12313-8_17
Young, R. J. "EQUITRIAGE: A Fairness Audit of Gender Bias in LLM-Based Emergency Department Triage."
Young, R. J., & Moody, G. D. "A Validated Prompt Bank for Malicious Code Generation: Separating Executable Weapons from Security Knowledge in 1,554 Consensus-Labeled Prompts."
Young, R. J. "Why Models Know But Don't Say: Chain-of-Thought Faithfulness Divergence Between Thinking Tokens and Answers in Open-Weight Reasoning Models."
Young, R. J., & Matthews, A. M. "CardioEmbed: Domain-Specialized Text Embeddings for Clinical Cardiology" & "Comparative Analysis of LoRA-Adapted Embedding Models for Clinical Cardiology Text Representation."
arXiv
2511.10930 · 2511.19739
Young, R. J. "Evaluating the Robustness of Large Language Model Safety Guardrails Against Adversarial Attacks."
Ghaharian, K., Soligo, M., Young, R., Golab, L., Kraus, S. W., & Wells, S. "Can Large Language Models Address Problem Gambling? Expert Insights from Gambling Treatment Professionals." + "Across the Bettor-Verse" (open banking, UK).
Wilkins, E. W., Young, R. J., Davidson, R., et al. "The influence of transcranial alternating current stimulation on the excitability of the unstimulated contralateral primary motor cortex."
Gada, E., Pangburn, P., Sahr, C., Schaben, C., & Young, R. "Unfulfilled Home Health Referrals Lead to Higher Mortality Among Medicare Advantage Members."
Zhuang, X., Young, R., Tillett, R., Cordes, D., & Oh, E. "Development of an AI approach that employs genomic and brain-imaging features to improve the diagnosis of Alzheimer's Disease."
Lyon, K. F., Cai, X., Young, R. J., et al. "Minimotif Miner 4: a million peptide minimotifs and counting." + "Minimotifs dysfunction is pervasive in neurodegenerative disorders" (Alz. & Dementia TRCI, 2018).
Lyon, K. F., Strong, C. L., Schooler, S. G., Young, R. J., et al. "Natural variability of minimotifs in 1092 people indicates that minimotifs are targets of evolution."
20+ abstracts and presentations across neuroscience, healthcare analytics, and AI venues.
2025
Clinton E, Young RJ, Matthews A & Poston B. "Enhancing clinical trial data extraction using GPT-4, Claude 3.5, and Llama 3.1 in Parkinson's disease." Society for Neuroscience, Washington DC, Nov 2025.
Davidson R, Poston B, Alhwayek G, Krider R, Young RJ, Clinton E, Park J & Karavadia S. "The Influence of Dual-Source Cerebellar tDCS on Muscle Fatigue." AAPM&R Annual Assembly, Salt Lake City, Oct 2025.
Davidson R, Krider R, Clinton E, Alhwayek G, Young RJ, Park J, Mouradian J, Karavadia S & Poston B. "The Effect of tDCS of the Dorsolateral Prefrontal Cortex on Muscle Fatigue Resistance." AAPM&R Annual Assembly, Salt Lake City, Oct 2025.
Young RJ, Ghaharian K, Golab L, Kraus S, Wells S & Soligo M. "Comparing LLM and Human Expert Responses to Problem Gambling Questions." Springer AIR-RES'25.
2024
Clinton E, Chauhan AS, Young RJ, Wilkins EW, Pamaran J, Karavadia SD, Riley ZA & Poston B. "Bilateral dual-source tDCS of primary motor cortex on muscle-fatigue progression." Society for Neuroscience, Washington DC, Nov 2024.
Young R, Contini V, Clinton E, Wilkins EW, Park JA, Mouradian J, Riley ZA & Poston B. "Dual-source premotor cortex tDCS on muscle fatigue in hand muscles." Society for Neuroscience, Washington DC, Nov 2024.
Poston B, Wilkins EW, Davidson R, Krider R, Alhwayek G, Shihady F, Yi K & Young RJ. "tACS application on contralateral primary motor cortex excitability." Society for Neuroscience, Washington DC, Nov 2024.
Li H, Chauhan AS, Clinton E, Wilkins EW, Young RJ, Park JA, Karavadia SD & Poston B. "Bihemispheric tDCS of primary motor cortex on muscle fatigue." AAPM&R Annual Assembly, San Diego, Nov 2024.
Young R & Zirath G. "Longitudinal Analysis of Medication Switches and Dosage Changes Among GLP-1 Receptor Agonists." Optum Partnership Forum, Dallas, Apr 2024.
Young R & Zirath G. "Effects of Discontinued Use of GLP-1 Receptor Agonists on Glycemic Control, Body Weight, and Lipid Profile." Optum Partnership Forum, Dallas, Apr 2024.
Young R & Zirath G. "Expanding the Therapeutic Spectrum: GLP-1 Agonists for Weight Management in Non-Diabetic Patients." Optum Partnership Forum, Dallas, Apr 2024.
2023 & EARLIER
Wilkins EW, Kawana E, Lopez Mora E, Houston D, Young RJ, Boss S, Riley ZA & Poston B. "tDCS on contralateral primary motor cortex excitability." Society for Neuroscience, Washington DC, Nov 2023.
Zhuang X, Tillett R, Young R, Cordes D & Oh E. "Associations between polygenic hazard score and Alzheimer's Disease progression and biomarkers by ApoE genotype and sex." Society for Neuroscience, 2021.
Young RJ, Lyon K, Oh E & Schiller MR. "Minimotifs in Human Traits and Disease." NIPM Symposium, Las Vegas, May 2018.
Tejada MB, Young R, Mercado V, Kaur K, Ikeda A & Lukkahatai N. "Association of telomeres and telomerase with clinical outcomes in children with acute lymphoblastic leukemia." Oncology Nursing Forum 43(2), 2016.
Wells S, Golab L, Young R & Ghaharian K. "BetterBettor: Promoting Responsible Gambling and Sport-Betting Knowledge Through an AI Chatbot."
Peterson J, Ghaharian K & Young R. "Beyond the Bet: An Open Banking Perspective on Gambling in the UK."
06
Books
The Neuroscience of Artificial Intelligence: A Journey into the Minds of Machines and Humans
27-chapter textbook on the bidirectional relationship between computational neuroscience and AI · neural coding, mathematical frameworks, modern AI systems, frontier topics. Integrates 100+ papers as annotated deep dives; 27-lab computational workbook.
2025
Healthcare Analytics and AI: Building Systems That Actually Work
22-chapter graduate textbook spanning the full stack of healthcare AI · privacy engineering, supervised learning, medical imaging, clinical NLP, LLMs, agentic workflows, bias, and governance. Integrates the author's research with 764 annotated papers.
2026
Data Engineering with Alteryx · Houghton, P. (Packt)
366-page guide to DataOps with the Alteryx stack. Senior Technical Reviewer.
2022
ISBN 9781803236483
07
Impact in Practice
AI-Driven Member Outreach at UnitedHealthcare
As the senior AI scientist behind the MLOps distress-scoring and member-identification pipelines for the Consumer Resolution Center, I surface at-risk members daily for human outreach teams to act on. In its first year the program served 150,000 distressed members, lifted satisfaction from 0 to 8.8/10, and scaled to 5M+ customers, recouping a $20 to 25M investment many times over. Documented in Harvard Business Review (2026).
150K
members served, year one
5M+
customers scaled to
Harvard Business Review, 2026
Personalized-Medicine Supercomputer, UNLV
Designed and deployed the Nevada Institute of Personalized Medicine's genomic HPC cluster with Advanced HPC and Bright Cluster Manager: single-pane management, cloud bursting, and reproducible image management for large-scale genomic analysis. Profiled in Bright Computing's launch press release and customer case study.
2017
Bright Computing press release & case study
PART III
Applied & Production
08
Production AI Work
Conversational-analytics and agentic systems on Databricks for UnitedHealth Group / Optum. Five production platforms in flight (May 2026); ~150 deduplicated LLM-orchestrated tools; largest deployment serves 47.5M member records plus 26.4M vector-indexed call transcripts via Databricks Apps with SSO.
CxS Insights · consumer abrasion platform
Serves Customer Experience Services teams at Optum / UHC via SSO · querying member distress, abrasion, drug issues, and call-center metrics without writing SQL. Deterministic Router → Planner → Executor; hybrid Haiku / Sonnet / Opus routing for ~40% cost reduction. 53 LLM-routed tools; Vector Search with reranker on 26.4M transcripts.
May 2026 · active
47.5M
member records served
26.4M
transcripts indexed
Ask Thaur · call understanding platform
Conversational SQL over 150M+ UHG / Optum call-center records; Claude Opus translates English to SQL via 21 tools with a 3-agent validation pipeline. Replaced a 2–3-FTE workflow ($5–6K/mo) with a self-serve UI at ~$100/mo.
Call-analytics agent · voice of consumer
~50×
cost reduction
60×
latency improvement
Ask Richard · upskilling platform
Used daily by 80+ Optum / UHG data scientists as the team's source of truth on Unity Catalog, Delta Lake, and SQL/ML. 56-reference corpus indexed as 27,982 documents + 6,687 images; 3-mode RAG dial; 12+ LLM ecosystem with a Claude proxy and 30+-file-type multimodal pipeline. Built on the Ask Lucky proof of concept (41.5M profiler records).
LLM with vs. without tool calling · built for MIS 769
Plus 6 more Spaces (IS 335 analytics demo, image-analysis tools, and others) · huggingface.co/richardyoung
10
Patents & Intellectual Property
Systems and Methods for Determining Readmission Rates
A computer-implemented, AI-based method for determining disease-specific readmission rates from hospital admission data; ML clusters the data and predicts readmission rates, output via a GUI.
U.S. Appl. 18/323,518
2023
Advanced ML System with Dynamic Adaptive Learning for Healthcare
Proprietary AI methods with adaptive learning and explainable-AI techniques (SHAP, LIME) for clinical decision support · transparency, auditability, and model updates under distribution shift. Confidential pending filing.
In preparation
2026
11
Grants & Funding
LLM Safety Research
OpenAI
$5,000
Jan–Jun 2025
RAG & Knowledge-Bot Design
Amazon Web Services
$20,000
Nov 2024–Oct 2025
DEC for problem-gambling risk identification
Nevada Council for Problem Gambling
$4,000
2024–2025
Minimotif & GWAS Bioinformatics
Private / Amazon Web Services
$10,000
2017–2018
G-RISE at UNLV · Multi-College NIH T-32
NIH T-32 Graduate Research Training Initiative. Role: Operations Director. Coordinated the multi-college submission; led data acquisition through the UNLV Office of Decision Support; secured dean-level letters of support; compiled materials across hundreds of hours with the PI and PD/PIs.
PI: E. Abel-Santos · 1T32GM161968
$2.22M
awarded · NIH T-32 (2025)
PART IV
Education & Experience
12
Education
Ph.D., Neuroscience
Interdisciplinary Neuroscience, Graduate College · University of Nevada, Las Vegas
Thesis: Machine Learning Methods and Transcranial Direct Current Stimulation for the Understanding and Treatment of Parkinson's Disease.
2025
Advisor: Brach Poston, Ph.D.
GPA 3.96
M.S., Data Analytics
Department of Computer Science · University of Nevada, Las Vegas
Thesis: Parkinson's Disease, tDCS, and Clinical Trials · using Machine Learning and LLMs to accelerate clinical-trial research.
2023
Advisor: Jorge Fonseca, Ph.D.
GPA 3.94
B.A., Psychology · magna cum laude
Honors College · University of Nevada, Las Vegas · Minors: Information Technology, Neuroscience, Honors
Honors Thesis: Minimotifs in Human Traits and Disease.
2017
Advisor: Martin Schiller, Ph.D.
GPA 3.6
13
Experience
INDUSTRY
Senior AI Research Scientist
UnitedHealth Group · Remote (Las Vegas, NV)
Design and ship production conversational-analytics and agentic AI on Databricks (5 platforms; ~150 LLM-orchestrated tools). Built the MLOps distress-scoring pipelines behind the Consumer Resolution Center featured in Harvard Business Review. Lead adversarial safety and instruction-adherence evaluation of frontier models for patient-facing deployment.
2024–Present
Senior Data Scientist
OptumCare · Las Vegas, NV
Built and deployed clinical and member-analytics ML across Medicare Advantage populations: readmission prediction (basis of a U.S. patent application), GLP-1 medication-outcome studies, and home-health referral / mortality analyses published in AJMC. Delivered 64 enterprise Tableau dashboards across operations, quality, and revenue cycle; led federal data submissions for a multi-billion-dollar merger (FTC/DOJ, 10+ departments); predictive-scheduling models freed 5 FTEs and delivered $1M+ in validated annual savings.
2018–2024
ACADEMIC
Machine Learning Researcher · Open Banking Data Research
UNLV International Gaming Institute (IGI)
2024–2027
Assistant Professor-in-Residence
Information Systems · Lee Business School · UNLV
2026–Present
Part-Time Professor
Lee Business School · UNLV
2023–2026
Machine Learning Research Assistant
Nevada Institute of Personalized Medicine
2020–Present
Lead Data Scientist
Nevada Institute of Personalized Medicine · designed and deployed the NIPM genomic HPC cluster (Slurm, cloud bursting, reproducible image management via Bright Cluster Manager), lifting compute throughput ~60× and enabling 100+ interdisciplinary research projects.
2014–2019
PART V
Teaching, Service & Recognition
14
Invited Talks
NATIONAL
John Snow Labs Applied AI Healthcare · Keynote to ~50,000+ data scientists: AI attack vectors in healthcare deployment.
2026
Concorde College Commencement · Keynote / Main Speaker.
2026
Applied Healthcare AI Summit · When the Safety Net Becomes the Attack Vector.
2026
Optum Innovation Network · AI for Physicians.
2025
Alteryx · Generative AI and the Future of Education · company-wide talk.
2024
REGIONAL · SELECTED
2026Mountain West CTR-IN (NIH IDeA) · AI Radar
2025Nathan Adelson Hospice · AI and End-of-Life Care
2025Optum · Nvidia Rapids & Databricks Photon
2024Nevada Academy of Family Physicians · AI/ML in Healthcare
2024President's Innovation Challenge · Generative AI
+8additional regional talks
15
Teaching Portfolio
Focus: AI/ML for Healthcare; LLM Evaluation & Safety; Applied Data Science & Business Analytics; Clinical Informatics. Graduate learners across UNLV MBA, Finance, and MIS tracks. Philosophy: teach logic, troubleshooting, and critical thinking as transferable skills via real-world projects and individualized video feedback.
"Those who know, do. Those who understand, teach."
— attributed to Aristotle
4.86/5
Student learning rating
13 to 19% above department 8 to 12% above UNLV benchmarks
vs. MIS dept 4.14 · Lee 4.32
COURSES · INSTRUCTOR OF RECORD (UNLV LEE BUSINESS SCHOOL)
IS 335 Business Analytics (UG) · 4 terms, 60–62 students/section.
'24–'26
MIS 776 Business Analytics (Grad) · total 4.89/5.
'23
MIS 769 Big Data, NLP & LLMs (Grad), 3 terms · 4.78/5.
'24–'26
MIS 752 Advanced Healthcare Analytics & ML (Grad) · scheduled, built around the author's textbook.
'26
Mentoring: Keith Webster (Independent Study, MS Applied Economics; President, Global BI, JD Power); President's Innovation Challenge faculty mentor, 5 students.
16
Professional Service
UnitedHealth Group Enterprise IRB
Co-Chair (2026–Present); Scientific Member, Generative AI & Big Data Domain Expert (2021–Present). Scientific oversight of human-subjects research protocols spanning the full UnitedHealth Group and Optum enterprise: 70,000+ affiliated physicians and the many integrated and non-integrated care organizations under UHG/Optum, including Reliant Medical Group, WellMed (Texas), and international affiliates.
Co-Chair
2026–Present
JOURNAL & PANEL REVIEW
American Journal of Managed Care (ML/AI, 2022–) · Brain Sciences (MDPI, 2024–) · Information Systems Journal (2025–) · Brain Research (2026–) · McGraw Hill (2024) · Concorde College Program Advisory Committee · Pediatric Tumor Board, Cure for the Kids / Sunrise Hospital (2015–16).
17
Honors & Affiliations
HONORS & AWARDS
2022UHG Emerging Leaders Program
2021Community-Based Participatory Research Award (UNLV)
2016Phi Kappa Phi · Nevada Org. of Nurse Leaders · Nevada Nurse Foundation Scholarship
2014Golden Key International Honor Society · Bob Davis Scholarship (2015)
1999Student of the Year Award
PROFESSIONAL ORGANIZATIONS
Society for Neuroscience Association for Information Systems American College of Medical Genetics International Neuroinformatics Coordinating Facility American Medical Informatics Association American Statistical Association American Society of Human Genetics
18
Certifications & Training
Generative AI FundamentalsDatabricks
AWS Technical Essentials · ArchitectingAWS
Business Intelligence ProfessionalGoogle
AI for Medical DiagnosisCoursera
Finetuning Transformers · NLPCodecademy
Alteryx SparkED Faculty AdvisorAlteryx
Plus extensive CITI / human-subjects research training (MIT, UHG, UNLV): AI & Human Subject Protections, IRB Members, Biomedical PI, GCP, Bioethics, Responsible Conduct of Research. All certifications meet NIH requirements for human subjects research training.
Richard J. Young · Curriculum Vitaecv.deepneuro.ai
