After 15 years of applying data science across a variety of different verticals, I've found that the combination of human + machine can be far more effective than either one by itself. I try to facilitate that through my work as a technologist and data scientist for early- and late-stage technology companies.
Building technology that is simple, stable, and scalable
Architecting and building a scalable platform can be a very difficult challenge, particularly in start-up companies where there are constantly trade-offs made to balance the need to ship software with the desire to avoid building up too much technical debt. These are the decisions that I thrive on. I've re-architected a machine learning platform from top to bottom within 6 months in order to simplify the code base, make it more modular, enhance software velocity, and improve scalability.
Building models that can automate repetitive tasks or enhance human decision making
The world of data science continues to march forward at an unbelievable pace with cutting edge tools and techniques emerging on a constant basis. I enjoy staying abreast of all these developments and leveraging the latest and greatest tools in order to build robust models that are capable of achieving a variety of goals:
- Causal Inference: As a trained economist, I began my career studying the nuances of causal inference and learning how to build robust models that are suitable for publication. I then transferred these skills to my work at Evolv and HiQ Labs, using a variety of indicators to predict employee attrition and performance on the job and publishing a number of studies showing that these relationships were indeed causal.
- Machine Learning: Over time, I shifted my skill set to pure prediction and the substrate on which I worked to the financial services sector where I spent years building models that predict fraud and default for large lenders and stood up an entire lending platform within 5 months.
- Natural Language Processing (NLP) and Generation (NLG): Working with unstructured text can be far more difficult than structured data but I enjoy sense making and extracting insights from it. For a company that I co-founded, RapportBoost.AI, I built models capable of ingesting millions of chat-based conversations, modeling them, predicting the message that was most likely to come next, and then optimizing that content via A/B testing.
Machine Learning Operations (MLOps)
Machine Learning Operations (MLOps)
Deploying machine learning models to a production environment in a scalable way
Model building is often much less challenging than deploying the models to a production environment, monitoring performance, and preventing model drift. Acquiring the skills to not only build the models but deploy them scalably is a rare capability that I have worked hard to develop. Much like software, the key is to achieve high velocity model building by leveraging robust pipelines that can automate as much as possible of the model building and deployment process. This facilitates continual releases (CICD style) to prevent model drift and model training in hours (not days or weeks like data scientists typically require).
From early in our journey, Mike helped us understand what was possible with our data, and to chart a course to use it--from data design to team-building. He has helped us expand the impact we have on small businesses.
CEO and Co-founder, Homebase
We engaged Mike on a very important project which involved analyzing multiple data sets to determine whether our platform served as an early-predictor of experience across several key domains. Mike was able to quickly outline a plan to attack our thesis and then worked hand-in-hand with our internal team, applying his expertise with machine learning and more. Mike is highly adept at both the technical side of a project as well as being able to meaningfully share his findings. Mike is an asset to any data strategy - personable, team oriented, and highly skilled; Mike is an outstanding resource for any tech company.