Project: Mocavo

Previous full-time employment from Jan 2014 to May 2016

Mocavo was a genealogy search engine that enabled users to search 6 billion records for free. At our peak, we were one of the largest site in the world on Google's index, with over 1 billion pages indexed. Mocavo was acquired by DC Thompson in the largest Techstars exit to date. I worked at Mocavo before, during, and after it's acquisition, which has been one of my most enriching life experiences. I learned so much and made a few friends along the way that I still talk to on a daily basis.

View project →

Data Science
Machine Learning
Laravel
PHP
jQuery

The business

Mocavo was a genealogy company that put billions of records on the public domain that either didn't exist on the internet, or could only be accessed by paying for subscriptions to sites like Ancestry.com. At Mocavo, we built a super fast website that optimized SEO and provided our users with free search of and access to all of our records. We eventually eclipsed over a billion pages in Google's index and funneled tens of thousands of unique users from organic search to our site each week. Because of our immense SEO value, Mocavo was acquired by DC Thompson in the largest Techstars exit to date.

My projects

While my official starting role was Data Scientist, a lot of what I did early on at Mocavo was learn how a tech company is run. I spent days with data analytics, days with engineering, days with marketing, days with design, etc. I soaked up so much knowledge in my first few months at Mocavo and had a blast while doing it. Everyone I worked with was patient and supportive of teaching me the ropes. Over the course of my time at Mocavo, I slowly transitioned from a Data Scientist role to a Software Engineer role. Most of my early projects were data analytics and creating models/machine learning algorithms to optimize various aspects of our business. I had an interest in web development and over time I started to both build the data models and then deploy them on our website as an end-to-end project.

Automated Adwords Bidder

One of my favorite projects while working at Mocavo was building an automated Adwords bidder that controlled a portfolio of Adwords keywords. For several years our Adwords account was managed by hand and usually driven by light analytics/statistics in Excel. I wrote a machine learning algorithm that learned to mimic these updates and take into account arbitrage opportunities for keywords based on the bids of our entire portfolio. I also wrote software that would automatically pull the Adwords data, make adjustments to bids, and upload the new bids every day. This project increased our Adwords efficiency (which we calculated using profit/click) by ~15%.

Executive Reports Dashboard

Our need for executive reports on traffic, subscribers, retention, etc. grew organically with the company. In the early days, a few SQL queries was all that was needed to generate reports for the executive team to track progress. As we grew and approached acquisition, our needs for a more streamlined and in-depth solution grew. I was tasked with creating a web-based dashboard that aggregated all the data shown in the executive reports. At this point, the manual reporting was taking up a few hours per week. By putting the data in a web app we were not only able to replicate the statistics our board had grown accustom to seeing, we were able to add graphs, tables, and other visualizations that made the data both more interpretable and more interactive.

Customer Behavior Predictive Model

Using a massive amount of data from Google BigQuery, a coworker and I shaped user data into time series that we fed into machine learning algorithms that learned to predict if a user was about to leave the site, unsubscribe, purchase a subscription, etc. Using this data, we were able to deploy a production analytics system that tracked users real-time and predicted with ~80% accuracy if they were about to initiate a "critical action." We used this system to change the site based on user behavior, which was successful, though not as successful as we hoped. We didn't get a chance to tune this project long-term, and I think that if we did we would have figured out how to use these predictions to trigger emails, coupons, popups, etc. that optimized the retention and subscription rates of the site.