Evan Curtin

Machine Learning Scientist/Engineer and former Computational Chemist

Chicago, IL | evanmcurtin@gmail.com | github.com/ecurtin2

SKILLS

7+ Years: Python (numpy, scipy, pandas, etc), Scientific / Mathematical Computing
4 Years:git, SQL, Linux/bash, Cloud Computing (Amazon Web Services), Docker
2 Years: Kubernetes, Argo, Apache Spark, C++
Interested in: Rust, Scala, Elm, Golang

Created a fully automated machine learning model refit pipeline on Kubernetes with Argo. The pipeline transforms over a Terabyte of raw data and results in a production-ready model library being trained on new data and published daily. This prevents about $14M loss annualized.
Scaled data transformation and analysis using Spark, Kubernetes and vendor platforms. The pipeline autoscales infrastructure on-demand and achieves >= 90% resource utilization while active and almost none during downtime.
Optimized model performance using hyperparameter tuning (hyperopt) and objective weighting.
Created CI/CD for the model training pipeline, enabling continuous iterative improvement. The pipeline is deployed using the GitOps pattern and can be upgraded or rolled back in seconds.
Worked closely with business and platform partners facilitate daily model deployment.

Created a risk model for insurance claims using XGBoost to triage adjustor resources.
Reduced model training time by 90% by parallelization with Dask and code optimizations.
Greatly Accelerated model deployment to production by porting to AWS SageMaker and Lambda.
Exposed machine learning services as serverless API’s.

Researched quantum chemistry using high performance, parallel C++ on our supercomputer.
Developed methods in electronic structure theory to determine when spin symmetries break in the electron gas, which required diagonalization of very large matrices using the SLEPc high performance linear algebra package.
Rated as “excellent” by students as TA for Physical Chemistry and Statistical Mechanics.

Measured antibody therapeutic binding kinetics via Surface Plasmon Resonance. Kinetics & Affinity of drug candididates were used as part of candidate selection for autoimmune disorders and cancer.

Left candidacy for data science: Stability of Hartree-Fock Equations in symmetry-breaking solids

Thesis: Low Dimensional Models for Predicting Nanomaterial Properties

Evan Curtin and Karl Sohlberg. Quantifying electron transit in donor-bridge-acceptor systems using probabilistic confidence. Journal of Theoretical and Computational Chemistry, 2018

Quantized: Python Quantum Mechanics Library https://github.com/ecurtin2/quantized
I love to cook and joined the sourdough bandwagon during covid-19 quarantine.
I fall asleep to cppcon / pycon youtube videos.