Standing up a data science function at later stage startups often fails: here’s a framework for success

One of my summer goals was to explore some new data science and machine learning technologies, which definitely would need to include all the innovation happening on the deployment side of production machine learning models. A few years ago, I had the experience of running my models on EC2 instances, which was not an ideal experience. So I took a look at AWS Sagemaker, which allows you to build and deploy machine learning models and they also have some good documentation filled with examples.

In the past year I have come to appreciate the rollercoaster of health and especially how it relates to the people I care about. All of that is a story for another day, but building a data-driven foundation and weaving it into the DNA of a company is my background.

I have been looking for a fun machine learning computer vision use case so I can play around with a CNN implementation. I haven’t worked on too many deep learning problems and the technology has evolved so quickly. My wife and I don’t regularly organize and backup my iPhone pictures. As I was going through the thousands of pictures, I realized a common theme:

Welcome to adulthood. After three years of marriage, my friends definitely empathize, things got real and it was time to do some grown up things around our life plan. After working on Wall Street and several FinTech startups spanning personal finance, financial planning, mortgage, and property and casualty insurance…..embarrasingly I didn’t ever consider life insurance. My wife and I realized that we both should have probably gotten some earlier in our lives when things were much cheaper, but better late than never, right?

Around 3 years ago I was looking for a fun supervised learning problem where I could familiarize myself with the popular python machine learning library scikit-learn. I found a dataset on Kaggle that had Lending Club’s historical default rates on 2014 and 2015 vintages and thought that it may be a fun exercise to apply some machine learning methods and see if I can beat the performance of a portfolio I constructed manually.

First post in 8 years! A lot has changed since then both macroeconomically and for myself professionally since my wordpress blog Extremely Old Blog

Reposted 2019 from Legacy Site