Kevin Spiekermann

Applied Scientist @ Amazon AGI | Previously ML Scientist @ Merck | PhD @ MIT

Greater Seattle Area

About

Machine learning scientist experienced with traditional and deep ML architectures, predictive and generative models, and uncertainty quantification. Currently an Applied Scientist at Amazon AGI working on LLM agents and GenAI at scale. Also have experience with code development, cheminformatics, and developing ML models to both predict chemical properties and generate structures. I enjoy collaborating with others to drive scientific and business impact. https://scholar.google.com/citations?hl=en&user=qg2LmbgAAAAJ

Experience

  • Applied Scientist II at Amazon
    Nov 2024 - Present · 1 yr 9 mos

    Generative AI – LLMs for Alexa/AIDo

  • Senior Machine Learning Scientist at Merck
    Sep 2023 - Nov 2024 · 1 yr 3 mos

    • Designed virtual libraries using generative modeling and reinforcement learning. • Delivered models to predict ADME properties and accelerate DMTA cycles. • Evaluated uncertainty metrics to prioritize compounds during active learning iterations. • Let efforts to democratize GenAI tools across modeling department.

  • PhD Candidate in Chemical Engineering at Massachusetts Institute of Technology
    Aug 2018 - Aug 2023 · 5 yrs 1 mo

    Thesis Title: Enabling Accurate and High-Throughput Kinetics Predictions via Message Passing Neural Networks Sample projects: • Developed and tuned a graph convolutional network (GCN) to accurately predict reaction barrier heights • Created new kinetics dataset with high-quality CCSD(T)-F12 calculations • Created new Python software to automate kinetic model construction and refinement • Substantially improved estimates for thermodynamic and kinetic parameters within RMG-database • Created kinetic model to understand pyrolysis of a surrogate mixture of heavy oil

  • Machine Learning Scientist at Microsoft
    Jan 2022 - Feb 2022 · 2 mos

    • Trained logistic regression and LightGBM models to predict income • Used SHAP values to analyze how unfairness mitigation impacts model structure • Contributed to the open source fairlearn repo

  • Engineering Consultant (MIT Practice School) at AstraZeneca
    Mar 2021 - May 2021 · 3 mos

    • Led team in training a linear latent variable regression model to predict species concentration from spectra • Analyzed the model’s physical interpretability and created internal tooling for reproducibility