Michael Byrd Email & Phone Number

New York, Ny, Us

San Francisco, California, Us

San Francisco, California, Us

As a member of the Ecosystem team during Reddit's path to public readiness, my primary focus involves solidifying analysis and modeling efforts around metrics shared throughout different product, finance, and marketing teams. While also helping mature the organization's data efforts with a successful IPO in mind.In particular, I am most focused on building pipelines to standardize metric forecasts for external, board, and executive level decision makers, all of which requires much cross-functional efforts between engineers, data scientists, and business leaders.● Designed, socialized, and built a full end-to-end configuration driven forecasting framework using Airflow, Kubeflow, BigQuery, and Python. This work involved collaboration between different Data and Infra teams, as well as gaining buy-in and trust from data scientists and product teams.● Worked with cross-functional partners and leaders to build and incorporate forecasts into key business models throughout Reddit. Examples include required models for public readiness (i.e. necessary for IPO), important internal financial models to understand expected revenue, and general team KPI goal setting.● Developed a cost optimization tool for AWS compute reservations in collaboration with finance and infrastructure teams.

Louisville, Ky, Us

After continual delivery and recognized thought leadership in the organization, I was promoted to a technical lead role. As a Lead Data Scientist, I provided technical leadership for a team of four, which consisted of designing products, outlining implementations, and managing customer and stakeholder expectations.● Mentored junior and senior data scientists on how to best make continuous impact with data and models, while also working toward long term goals and ensuring maintainability for the organization.● Launched a site selection product to identify strategic, high impact locations across multiple countries to open new stores. Used Snowflake's and PostgreSQL's GIS functionality, an ensemble of various machine learning models, and geo-clustering with DBSCAN in Python.● Supported front-end engineering efforts to ensure necessary deadlines were met and to guide product design to best enable the necessary machine learning model outputs.

Louisville, Ky, Us

I was brought on to support developing machine learning based applications for Yum! as part of their, at the time, new and growing data organization. This work was mainly aimed at improving how Yum's brands (KFC, Taco Bell, and Pizza Hut) operated, and involved collaborating with different brands' operations and marketing teams.● Ushered the development of a marketing campaign tool to improve objectivity and efficiency of marketing experiments. This included test/control store recommendations with Bayesian causal analysis for evaluating experiment outcomes.● Designed and implemented a library agnostic forecasting service in Python for rapid prototyping and deployment.● Observed a 10% overall improvement to accuracy with a 7x faster computation time when forecasting millions of products a day after the introduction of RNN and Transformer deep learning architectures.

Southlake, Texas, Us

As a senior data scientist in Sabre Research, I focused on R&D work through several different teams at the company. This work mainly comprised of implementing new models for hard problems that could not be solved traditionally, and providing consultation on statistical tasks.● Created a pipeline with Python and TensorFlow 2 to learn customer preference of itineraries at scale (~TBs) with SOTA neural networks.● Identified weaknesses in system architecture that prohibited AI/ML enhancements. Contributed to the redesign of those processes and built business cases for the value of each component’s enhancement.● Developed a proof-of-concept to identify quality usage of a Sabre product that scaled as was never before possible between several product features, organizations, and agents.● Consulted on a complex experiment targeted at better diversifying product offerings, where I reduced the number of feasible settings from 264 to 4 with offline analysis.

Southlake, Texas, Us

As a Data Scientist in the Sabre Research org. I performed various interesting analysis around different online experimentation practices. Many of the topics involved statistically correct continuous testing and real-time learning with various bandit algorithms.● Consulted with a new experimentation platform to run power analysis for complex testing scenarios.● Designed routines for identifying unique sets of parameters that provided incremental add to be used in experiments to improve conversion.● Investigated a proof-of-concept for a general real-time learning system using contextual bandits, illustrating an exponential improvement in learning with the same amount of data.● Collaborated to help build a real-time A/B/n testing platform using Thompson sampling and batch racing bandit algorithms for determining optimal settings for driving up conversion.● Developed interpretable forecasting models with seasonal ARIMA and Bayesian state space models for KPIs that are used by decision makers for planning future resource allocation

Southlake, Texas, Us

● Improved the predictive performance of a classifier used for process automation by 25% with XGBoost implemented from R.● Implemented a neural network with convolutional layers (CNNs) via Keras for a large scale predictive task involving natural language (NLP).● Used a penalized Cox proportional hazards model to determine the best length of time to cache a set of flight prices given characteristics of the itineraries.

Dallas, Tx, Us

As a graduate teaching assistant, I ran statistics labs and helped students better understand statistical concepts in realm of their disciplines.● Led statistics labs to reinforce course material with Excel implementations.● Worked with small groups of students to better facilitate understanding of statistical methodology

Dallas, Tx, Us

As a statistics lecturer, I had the opportunity to lead two of my own intro statistics courses at SMU. My guiding principles here were to focus on sound conceptual understanding, leaving calculation to a minimum where possible.● Developed skills to present statistical methodologies to small and large groups of individuals whom are not familiar with statistical practice● Taught students how to quantify and give data backed solutions to practical problems with focus on proper technique, presentation, and implementation.

Fort Worth, Texas, Us

● Researched and developed optimal routines for auto loan approval classification models.● Improved predictive accuracy for sub-prime lending by 5% with implementation of a new dimension reduction technique.● Used SAS and SQL to access large scale databases for model building.

Doctor Of Philosophy - Phd, Statistical Science

Bachelor Of Science - Bs, Economics