Greg Crysler Email & Phone Number

Austin, Tx, Us

Austin, Texas Metropolitan Area

process.begin()process.persuade()process.prove(assumptions = GoodFaith)process.multiprove()#from Batman import Intimidate

Austin, Texas Metropolitan Area

● Regional and national econometric models and data analysis to identify underlying macro conditions (e.g. consumer spending, inflation, housing, etc.) affecting business performance.● Responsible for production model scorecards and packaging forecast information for corporate officers and board of directors.● Google Cloud Platform engineering and solution design (e.g pipelines, maintaining model registry, migrating VMs, stored procedures, etc.)

Austin, Texas Metropolitan Area

● Implemented an AWS-based PyTorch ML pipeline utilizing CUDA (GPU processing) to forecast long-range expected Nielsen Ratings (i.e. supply-side), expected market value of commercial inventory and expected sold units (i.e. demand-side). Components of the pipeline include:• NBEATS (Recurrent Neural Network, “RNN”) from the DARTS library for forecasting TV Ad Sales. NBEATS maximizes interpretability (2 layers: 1 for Trend, 1 for Seasonality) with the appropriate balance of high precision.• TRANSFORMER from the DARTS library to model 1 Supply Series (Nielsen Ratings) + 1 Demand Series (Average Commercial Unit Price) concurrently. • Used to support senior leadership’s financial / cash-flow planning.

Austin, Texas Area

Sold $2.6m in work (conservatively)● Implemented the budget forecasting platform utilized by the Department of Veterans Affairs’ (VA) to support the Mission Act (https://missionact.va.gov/)● The platform integrates 3 authorization and 5 claims processing systems across 6 servers and 11 independent schemas.● The forecasting system uses two algorithms to model the sequence of VA spending on purchased, outpatient medical care: • (1) a hierarchical time series regression system to predict national demand for care outside of the VA system. It predicts every possible type of outpatient care for all 129 VA offices in the US.• (2) using the demand predicted by (1) as an input, together with dozens of other predictors, an ensemble supply model (Multi-Task Elastic Net and Multi-Target Random Forest*) predicts the resulting claims payments over time. The system was implemented in May 2019 and assists in the management of ~$15b in annual spending.The VA has requested 3 additional forecasting systems for Inpatient, Contract Dialysis and Nursing Home services. New contract signed in June 2020.

Austin, Texas Area

● Data architect and senior data scientist for a financial management tool used by over 100 offices within Department of Veterans Affairs’ (VA). Designed to improve compliance with GAAP standards, in 2017, the system led to the re-allocation of over $300m in previously reserved funds. The system is now fully implemented, assisting with the management of over $10b in annual expenditure for all VA purchased care programs.● Created the back-end infrastructure (R & SSIS) that integrated data from 5 VA enterprise applications. Trained and deployed to production 3 machine learning models (xgboost and elasticnet) to estimate purchased care liabilities based on 60m claims and 5m authorizations for care. ● Created automated sampling package to evaluate business operations errors for VA’s purchase care dialysis program (R & SSIS). The package parameterized confidence level, margin of error and estimated failure rate, allowing VA executives to customize the level of review for the department’s audit programs.

Austin, Texas Area

● Created an association rules procedure (a priori algorithm) to mine large-scale survey of customer perceptions.Within a dataset containing 400 questions for 12,000 respondents, procedure detected 5 response combinations that motivated customer purchasing decisions in ~80% of cases. Findings from the study supported annual strategic and creative media planning. ● Created a framework to select an optimal basket of test markets for a digital pilot program. The process evaluated 100+ markets’ revenue growth and competitive conditions to U.S. averages across 60 measures of similarity (Minkowski Distances and Information Divergence). The selection formula maximized the statistical independence of the markets while minimizing their collective variance to the national totals. The selected markets were used for the media planning strategy.● Created an Extract/Transform/Load (ETL) process to integrate data sources representing 20 distinct media channels (e.g. National Broadcast, Cable, Syndication, etc.). The final dataset was used to identity the optimal mix of media spending allocations for ~12 large U.S. local markets.

Austin, Texas Area

● Led data integration and credit model development for online lender using MySQL, R and SAS. Project included creating an Extract/Transform/Load (ETL) process to load 1 terabyte of files from 5 credit bureau sources. The final implementation partitioned the data into 5 sub-populations, applied a custom L1 regularization path algorithm for each group, and produced granular diagnostic output across the entire path.● Developed User Activity DataMart. SSIS package automated activity logs ETL, generated a synthetic user key to track actions across source data and structured data for time series analysis.

Hartford, Connecticut Area

● Implementation of Stochastic Boosting (ADA.M1 Algorithm) to optimize flag-based auditing procedure applying over 30 filters. Boosted tree-based ensemble reduced time to output by 25 hours; a 50% increase in performance. Development of ensemble learning techniques, including Hierarchical Mixtures of Experts (HME).● Performed all Database Administration tasks for 3 terabyte pharmacy claims server. Responsibilities included all Extract/Transform/Load (ETL), development (T-SQL), maintenance (SSIS), ad hoc reporting, and tuning.● Design and development of case management data model and operational reporting solution, including ETL into Oracle DataMart. Data model successfully integrated client’s operational activity data with case history information and was based on a source of 230 dimensional tables.

United States

● Led migration of mainframe SAS/DB2 Encounter Reporting System (ERS) to SAS/Oracle EDW solution. Responsibilities included development of views, testing, SQL Development Standards Guide and Extract/Transform/Load (ETL) documentation. ● Delivered SAS OLAP cube to automate data profiling review and analysis. ● Led migration of SAS Hospital Administrative Profile System to SAS/Teradata EDW solution. Completed this project without access to the original data source.

Houston, Tx

● Re-engineered specialty benefits dashboard review process for national health plan. Objectives included replacing manual QA process with Multivariate Analysis of Variance (MANOVA)-based stored procedures.● Led statistical analysis of specialty benefits utilization. including custom metrics for radiation therapy, general utilization and provider quality. ● Using T-SQL and SQL Server Reporting Services (SSRS), created Ad Hoc and Scheduled Reports.

Bachelor'S Degree, History; Public Policy

Graduate Certificate, Applied Statistics, 3.7

Activities and Societies: Statistical Inference from Transactional Data SourcesMachine Learning, Multivariate Analysis, Time Series.

Juris Doctor, Law

Thesis: Review of Internal Method Modeling Compliance Evaluation within Basel International Settlement (BIS) Accords.

Diploma, Mathematics & Physical Sciences

Activities and Societies: Probability3rd in my class... for whatever that's worth