David Boren Email & Phone Number

Washington, Dc, Us

Designed Luigi-based distributed framework for machine-learning training and deployment on kubernetes. Worked closely with DS teamates to improve upon this structure, allowing for faster model iteration and much more bulletproof validation and model book-keeping. - Deployment Tech: Docker, GRPC, Flask, Argo - Algo Tech: Gradient boosted trees, Collaborative Filtering, Regularized logistic, Multi-Task NN Survival - Python Packages: Sklearn, Pandas, Numpy, Xgboost, LightGBM, LightFM, Pysurvival, Luigi, MLFlowSpearheaded continuous integration, testing, and deployment of ELT processes for Data-Warehousing. Worked closely with our skilled BI team to put together a git-controlled warehouse. - Warehousing Tech: Postgres, Snowflake – Delta-Deployment Tech: CircleCI, Airflow, Argo, DBT - Sources: S3, Postgres, Dynamodb, SFTP, Dropbox - Replication Tech: SQLAlchemy-Core, AWS DMS, Alooma, FivetranDesigned and owned low-latency AB and Feature-rollout deployment in a microservices context. Worked closely with other backend teams to take a more experimentally-minded approach to feature development. - Datastore Tech: DynamoDB, Redis - Server Tech: GRPC, Pubsub (via SQS and SNS) - Segmentation Tech: Optimizely Created dev-onboarding bootcamp taking devs from zero kubernetes/grpc experience to full backend microservice deployment. Covered the basics of dependency locking, docker containers, ci + testing, service creation + deployment, logging, and error handling. - Languages: Python, Ruby, GolangMost recently took over inventory team and performed a sweeping refactor to replace technical debt, moving from postgres to noSQL and adding image processing features (image classification, instance segmentation, superresolution) along the way. - Deprecated Tech: Postgres, Celery - New Tech: Dynamodb, AWS SQS/SNS, pytorch, detectron2

Chicago, Il, Us

Predictive Analytics: Implemented training of the following models for in-production use: • Extreme Gradient Boosting • Random Forests • Elastic Net Regression • Ensembles of the above models Implemented Elastic Map-Reduce distributed R training system: • Distributed hyper-parameter tuning • Interface to AWS using boto3 python package • Wrote flexible R-interface to manage and view emr job submission and monitoring • Setup a convenient logging system for monitoring remote builds, including a redis channel system for real-time log viewing Created a simple, flexible interface for automatic generation of partial dependence plots across all classifier types using the ggplot2 library, incorporating a sampled variation of the statistical definition of true partial dependence.Predictive Model Training Environment Stabilization: Implemented dockerized R container builds and CI testing using Jenkins and CircleCI, including: • Bash and python scripting to ensure flexible and fast container building • Caching of library dependency bundles at the branch and master levels • Setting up and maintaining a variety of Jenkins and CircleCI projects, including one to create master caches and ensure bundled environments can be built from scratch. Made major modifications to our custom package management system, including: • Topological dependency sort • Reducing the dependencies of the system itself and removing much of the ad-hoc package installations previously required. • Increasing determinism across training environments with improved package parsing and installation logic

• Developed and tested agent-based models for simulation of HIV transmission in South Africa with various biomedical and behavioral intervention scenarios.• Utilized supercomputer cluster for highly parallelized execution and calibration of this simulation work.• Statistical Analysis of data and simulation results including multi-objective-optimization• Created and analyzed emergency-preparedness scenarios for LA County Department of Health. • Implemented a competing hazards survival model to predict infection and clearance rates at varying levels of exposure.• Applied a large-scale sensitivity analysis of a complex antibiotic distribution scheduling system for emergency response planning• Analyzed high-throughput data for nanoparticle toxicity assays on yeast and E-Coli cultures.• Applied false discovery rate methodologies for identification of genes affecting survival in conditions with high concentrations of nanoparticles of varying size.

Doctor Of Philosophy (Ph.D.), Biostatistics

Master'S Degree, Biomedical Engineering

Bs, Bioengineering