Michael Rawson
Applied Mathematics · Statistics · Artificial Intelligence

Michael Rawson, Ph.D.

AI Applied Scientist, Amazon · Seattle, WA

Applied mathematician and AI scientist with 6+ years building, fine-tuning, and evaluating machine learning systems in production at Amazon, Microsoft, Meta, and PNNL. Work spans the full AI lifecycle — LLM and agentic system development, foundation model post-training, reinforcement learning, adversarial robustness, and rigorous statistical evaluation with uncertainty quantification.

Ph.D. Applied Math & Statistics, UMD ICML · ICLR · SPIE · JMM Reviewer: ICML, NeurIPS, AMLC
Contact Download CV LinkedIn
6+Years in production ML
Ph.D.Applied math & statistics
13Selected publications
3National labs & agencies
01

Research & Technical Areas

Generative AI & LLMs

LLM development & evaluation AI agents (MCP) Prompt engineering Context engineering RAG Fine-tuning & post-training Mixture-of-Experts transformers GPT / Phi HuggingFace

ML & Statistics

Statistical modeling & inference Experiment design / A/B testing Causal & Lipschitz-bound analysis Uncertainty quantification Reinforcement learning (deep Q, bandits) Anomaly detection Time series forecasting Pruning & optimization Drift detection & retraining

Speech & Multimodal

Speaker & speech recognition Adversarial robustness for speech Multimodal information fusion Hyperspectral imaging MRI signal processing

Engineering

PythonPyTorchTensorFlow C++CUDASQL SparkDockerAWS SLURM / HPCMATLAB Data pipelinesProduction monitoring

Responsible AI

Adversarial robustness Safety & security evaluation Model reliability AI governance

Community

Open source: PyGraphCurvature ↗ Open source: Petri Nets ↗ Reviewer: ICML, NeurIPS, MLADS, TAG, JMM, Amazon CSS/AMLC Organizer: JMM Microsoft Hackathon judge
02

Professional Experience

May 2025 – PresentSeattle, WA

AI Applied Scientist

Amazon

  • Build agentic AI experiences on the Model Context Protocol (MCP), designing tool-use workflows and interaction patterns for long-horizon logistics tasks.
  • Develop Mixture-of-Experts transformer models for hierarchical product customs classification supporting global trade, including fine-tuning and post-training to lift task-specific accuracy and enterprise readiness.
  • Curate and structure training and evaluation datasets for hierarchical classification, defining benchmarks and quality metrics used to track model improvement.
  • Led the Tinc Hackathon team on theft and anomaly detection, placing top 5 nationally.
  • Partner with engineering, product, and business stakeholders to define model success metrics and translate evaluation findings into roadmap decisions.
Apr 2024 – May 2025Seattle, WA

ML Research Engineer

Microsoft

  • Built end-to-end RAG and LLM solutions for enterprise cyber security and safety analysts using OpenAI GPT and frontier Phi models, covering retrieval strategy, prompt and system instruction design, and evaluation.
  • Developed automatic prompt differentiation methods for systematic prompt optimization; presented at the Microsoft Security AI Research (MSECAIR) Workshop.
  • Designed evaluation approaches for LLM output quality, reliability, and safety in a high-stakes security context, identifying failure modes and closing them through context engineering and retrieval improvements.
  • Maintained and retrained production NLP models under data and model drift, monitoring performance telemetry and triggering retraining cycles in PyTorch.
  • Communicated technical findings to engineering, product, and executive stakeholders; served as a Microsoft Hackathon judge.
Apr 2022 – Apr 2024Seattle, WA

Machine Learning Researcher

Pacific Northwest National Laboratory

  • Built deep learning systems for speaker and speech recognition, including adversarial evaluation to measure and improve model robustness.
  • Developed a knowledge and ontology system mapping semantic distance via a lattice metric, supporting retrieval and structured reasoning over technical corpora.
  • Trained, evaluated, tested, and deployed models across LLMs, graph/network transformers, anomaly detection, and time series forecasting on GPU, SLURM, and HPC infrastructure.
  • Published first-author and collaborative work at ICML TAG, ICLR Physics4ML, and SPIE on efficient model architectures, neural network pruning, and fast hyperspectral abundance estimation.
Jun 2020 – Apr 2022College Park, MD

Research Assistant

School of Medicine, University of Maryland

  • Developed high-dimensional MRI reconstruction methods robust to noisy and corrupted data, combining signal processing, mathematical modeling, optimization, and deep learning.
Sep 2020 – Sep 2021Adelphi, MD

Machine Learning Scientist Fellow

Army Research Lab

  • Designed and implemented a deep reinforcement learning system for multimodal information fusion across heterogeneous sensor sources, with uncertainty quantification to bound decision confidence.
May 2021 – Aug 2021Menlo Park, CA

Machine Learning Engineer Intern

Facebook (Meta)

  • Applied deep Q-learning and deep bandit reinforcement learning to Facebook Marketplace search ranking, improving simulated search performance by 40%.
  • Built offline simulation and evaluation infrastructure to compare ranking policies before production exposure. C++, Cython, Python.
Aug 2017 – May 2020College Park, MD

Research Assistant

Norbert Wiener Center, University of Maryland

  • Analyzed deep neural network stability and generalization using Lipschitz bounds; investigated NLP representations with topological data analysis. C++, CUDA, Python.
Feb 2014 – Apr 2016New York, NY

Software Engineer

Bloomberg LP

  • Modeled financial derivatives data with machine learning over Spark-based data pipelines; built volatility surface approximation and trading order execution/matching systems. C++, SQL, TDD, Agile.
02b

Earlier Roles

Intern Research Assistant, Earth Sciences — NASA Goddard Space Flight Center
Greenbelt, MD · Jun – Aug 2017
Research Assistant — Courant Institute, NYU
New York, NY · May 2016 – May 2017
Software Engineer Intern — Amazon.com
Seattle, WA · May – Aug 2013
03

Selected Publications & Talks

04

Education

Ph.D., Applied Mathematics and Statistics — University of Maryland, College Park
May 2022
M.S., Mathematics — Courant Institute of Mathematical Sciences, New York University
Aug 2017
B.S., Computer Science — Herbert Wertheim College of Engineering, University of Florida
Dec 2013
05

Contact

Direct

Documents