Devin White Machine Learning Researcher

Curriculum vitae

Research, engineering, evaluation, and technical leadership across intelligent systems.

Machine learning researcher with 5+ years of applied AI research experience, an M.S. in Artificial Intelligence with a 4.0 GPA, and end-to-end experience taking projects from new research questions to working code, experiments, publications, and presentations.

LinkedIn ↗

Profile

Machine learning researcher who combines research depth with practical implementation. Experience includes originating and leading RLHF research, building reward-model and policy-training pipelines, developing language-model agents and evaluation environments, designing controlled experiments, mentoring research teams, and communicating technical results through peer-reviewed publications and invited presentations.

8 Research publications
70+ Research citations
7+ Students mentored
4.0 M.S. AI GPA

Experience

Machine Learning Researcher (Fellow)

ORAU / Army Educational Outreach Program

  • Lead and contribute to applied AI research in RLHF, reward modeling, reinforcement learning, compact language models, intelligent agents, and evaluation.
  • Take projects through problem formulation, algorithm design, implementation, experiment planning, analysis, publication, open-source release, and technical presentation.
  • Develop end-to-end language-model alignment pipelines using reward modeling, parameter-efficient fine-tuning, and reinforcement-learning-based policy optimization.
  • Build agent frameworks, interactive evaluation environments, and benchmarking systems for studying model behavior, reliability, and real-time decision making.
  • Mentor junior researchers, assign and review technical work, coordinate experiments, and guide project decisions across collaborative research teams.

Graduate Research Assistant

University of Texas at San Antonio

  • Advanced Rating-Based Reinforcement Learning from preliminary research to an AAAI 2024 main-conference publication and master's thesis.
  • Designed reward-learning methods, implemented reinforcement-learning systems, analyzed model behavior, and developed controlled experiments using synthetic and human feedback.
  • Contributed to reinforcement-learning research for autonomous guidance and time-constrained intercept problems.
  • Supported research writing, presentation development, and mentoring of student collaborators.

Technical Laboratory Assistant II

University of Texas at San Antonio

  • Implemented Rating-Based Reinforcement Learning in the DeepMind Control Suite and conducted an IRB-approved human-subject study with 20 participants.
  • Built data collection, experiment, and analysis workflows for comparing direct ratings with pairwise preference feedback.
  • Developed a custom Gymnasium environment for optimal time-constrained intercept guidance.

Undergraduate Research Assistant / NSF REU

University of Texas at San Antonio

  • Helped originate the Rating-Based Reinforcement Learning research direction and led a four-student implementation team.
  • Developed early reinforcement-learning experiments, presented preliminary findings, and established the technical foundation for later publications.
  • Built experience across algorithm implementation, experimental design, team coordination, and applied AI research communication.

Research leadership

Project ownership

Originated or helped define research directions, selected technical approaches, built core systems, investigated results, and led projects through publication.

Technical decision making

Evaluated alternative losses, architectures, feedback formats, metrics, and experiment designs while balancing research rigor with compute and timeline constraints.

Mentorship

Mentored 7+ students by assessing baseline knowledge, explaining technical concepts, assigning scoped work, reviewing results, and helping collaborators grow into independent contributors.

Communication

Presented research to academic, government, and executive audiences and translated complex findings into papers, visuals, open-source systems, and actionable next steps.

Education

M.S. in Artificial Intelligence · GPA 4.0

University of Texas at San Antonio

Member of UTSA's first graduating M.S. Artificial Intelligence cohort. Thesis: Reinforcement Learning From Human Ratings .

B.S. in Electrical and Computer Engineering · Cum Laude

University of Texas at San Antonio

Research publications

  1. Too Big to Think: Capacity, Memorization, and Generalization in Pre-Trained Transformers. ICML 2025 TTODLer-FM Workshop, Oral.
  2. Multi-Task Reward Learning from Human Ratings. ICML 2025 Models of Human Feedback Workshop.
  3. Human-Inspired Multi-Level Reinforcement Learning. NeurIPS 2025 ARLET Workshop.
  4. Performance Optimization of Ratings-Based Reinforcement Learning. AAAI 2025 CAIHU Bridge.
  5. Atari-GPT: Benchmarking Multimodal Large Language Models as Low-Level Policies in Atari Games. AAAI 2025 KnowFM Workshop.
  6. Rating-Based Reinforcement Learning. AAAI 2024.
  7. Deep Reinforcement Learning-based Optimal Time-constrained Intercept Guidance. AIAA SciTech 2024, Guidance, Navigation, and Control.
  8. Rating-based Reinforcement Learning. ICML 2023 Workshop.
Publication details and contributions →

Selected technical projects

Small-LLM RLHF Pipeline

End-to-end reward modeling, PEFT/LoRA fine-tuning, and RLOO policy optimization for compact language models with Hugging Face TRL.

Minimal LLM Agent Framework

Independent ReAct-style agent architecture with tools, retries, reflection, structured traces, and SQLite/FTS5-backed memory.

Atari-GPT

Multimodal language-model benchmark and interactive environments for studying low-level control, grounding, and action consistency.

ASCII Breakout MLX

On-device compact language-model agent implemented with Apple MLX for efficient inference on Apple Silicon.

RL-Driven nanoGPT

GPT-style policy trained from scratch with PPO in a custom ASCII game environment.

Simplified RbRL & PbRL

Readable research implementations for comparing direct ratings and pairwise preferences in reward learning.

View all projects →

Invited presentations & service

  • DEVCOM VIP Day: invited solo presentation on Atari-GPT and Rating-Based Reinforcement Learning.
  • ARL Technical Advisory Board: selected presentation on Rating-Based Reinforcement Learning.
  • AEOP / ORAU: invited research presentation covering applied AI work and relevance to Army research priorities.
  • IEEE Transactions on Cybernetics: peer reviewer.

Recognition & leadership

  • ICML 2025 workshop oral presentation for Too Big to Think.
  • Featured by UTSA as a member of the first M.S. Artificial Intelligence graduating cohort.
  • First Place, UTSA Tech Symposium.
  • Outstanding Undergraduate Award for Excellence in STEM.
  • Eagle Scout.