Leave The House Out of It
To call Leave the House Out of It my side hustle would misrepresent its importance to me. It’s a business that’s only ever had a dozen customers and has never been profitable, but it does keep my hands dirty with technology. I began my journey as a manager in 2013 at Fannie Mae and watched my contribution to the code base fall to nearly nothing as I took on more of a role as coach, mentor, organizer, and servant-leader to the development team. Not wanting to lose my technical skills, I invented an app that allows my friends and me to bet on football games against each other, and it keeps track of the score. There’s a more detailed explanation of how the idea formed here.
Like any good technology nerd, the project has been more about the joy of coding something than anything else. You can tell this by the number of times it’s been completely rewritten for no purpose other than to gain hands-on experience in new technology!
2014
It started in 2014 as a Google AppEngine project with data objects for its only storage.
2016
In 2016 (not surprisingly, at the same time when Fannie Mae asked me to roll out a new CI/CD process), I added a CI/CD pipeline in Jenkins, leveraged automated BDD testing from cucumber, and performance tests from JMeter.
2018
In 2018 (not surprisingly, when I started leading RBC’s Kubernetes implementation), the project moved to AWS. It was rewritten as a set of Spring-based microservices deployed on an EKS cluster with Helm and that same Jenkins pipeline.
2019
In 2019 (not surprisingly at the same time that I took over IBM GTS’s AWS Consulting practice it transformed into an entirely serverless application. The backend became Lambda processes driven by CloudWatch events, APIs, and each other. The database was converted from a mySQL relational database to a DynamoDB database.
2020
In 2020 (not surprisingly, while I was preparing for my AWS Professional Architect certification), I incorporated a bunch of AWS services, including tighter integration with CloudWatch for anomaly detection and a new CI/CD process to replace Jenkins.
2021
In 2021 (not surprisingly, at the same time I was moving into my role running the Apps and Data practice), I created a new player for Leave the House Out of It, Book-E. It picks games based on artificial intelligence. I outlined the creation process in three blog posts:
Machine Learning Holiday Project Part I
2025
In 2025 I revamped the user interface and modified the algorithm for calculating scores. The key here wasn’t WHAT I did, but HOW I did it. I worked with Anthropic’s Claude Code and learned a lot about what is possible with AI assisted coding. There are a bunch of reflections linked off of this post:
2026
In 2026 (not surprisingly, with Comites.ai already living on Google Cloud), Leave the House Out of It left AWS. The Java Lambdas and DynamoDB became a TypeScript service on Cloud Run, the web app moved to Firebase Hosting with Google sign-in, Cloud Scheduler and Cloud Tasks took over processing the games, and the whole thing is now Terraform, deployed by GitHub Actions. The UI got its first real redesign, too. It looks like the board at a Vegas sportsbook now, dot-matrix type and all, and it finally works on a phone. Along the way it picked up email invitations, weekly rosters, game pages with weather, injuries, and sportsbook odds, a League History tab with AI-written season recaps, and a public API so that bots can play.
Almost all of it was written with Claude Code, and the difference from a year earlier is the real story of the post: the intern grew up. Vibe Coding Update: It’s Gotten a Lot Better
2026: BOOK-E 2.0
The same year, Book-E got a rebuild of his own. The 2021 version was a Jupyter notebook and a SageMaker model that needed me in the loop every week. BOOK-E 2.0 is a small platform on Google Cloud built in three tiers: a BigQuery warehouse fed by MySportsFeeds, nflverse, and LTHOI itself (about 2,700 games, 70,000 player stat lines, and 68,000 timestamped line movements so far); a set of pluggable prediction engines, including two boosted-tree models trained in BigQuery ML, one on football fundamentals and one that also watches how the Vegas line moves, plus a baseline to keep them honest; and an agent that wakes up every couple of hours during the betting window, compares the models to the league’s lines, and only bets when he sees at least a two-point edge on a spread or three on a total.
Every bet and every pass lands in an audit log, the models retrain themselves each Tuesday, and the whole thing costs a few dollars a month. He is playing in The Originals this season with no human in the loop, and I am deliberately not looking at the standings until they mean something. BOOK-E 2.0: Rebuilding My Robot Gambler