A few years ago, a good friend and I built a side project to predict the outcomes of MLB games. Why MLB? Well, compared to other sports, baseball is played in a controlled atmosphere. Each at-bat starts the same way, it's effectively 1-on-1 between a batter and a pitcher, and there's a surplus of advanced stats that represent team defense, pitcher efficiency, etc that enable America's pastime to be simulated through numbers better than most other sports. Basically, the variables are controlled and predictable.
Plus, the two of us are nerds at heart who just love sports and numbers.
What We Did
We created a backend which scrapes daily matchups and up-to-date stats for the starting pitchers and the teams playing. These numbers are crunched (with a secret formula), and the expected results of the games are saved.
Tech Deep Dive
This project is written in Python and hosted on an AWS Lambda function which is triggered every day during the MLB season with a CloudWatch cron Event. The results are saved in both S3 and DynamoDB, and the results are made available through a private REST API hosted on API Gateway. The infrastucture is managed entirely with Terraform.
Since (this) blog site is hosted on GitHub Pages, and I don't want to push the API key to public source code, here's an example of the output results:
{
"away_win_perc": {
"S": "55.39%"
},
"date": {
"S": "2025-07-20"
},
"home_win_perc": {
"S": "44.61%"
},
"home_team": {
"S": "CHC"
},
"away_team": {
"S": "BOS"
},
"ttl": {
"N": "1816084868"
},
"total_runs": {
"S": "8.14"
}
},Improvements Moving Forward
As it always goes with software, there's plenty of improvements to be made that require refactors. We first started this project when we were out of college 5 years ago. Since then, I'd like to think I've become a significantly better engineer, and I can spot a lot of improvements I'd make if I did it all over again. Off the top of my head, here's a few examples:
- Better error handling
- Code cleanliness
- More separation of concerns between controllers, business logic, DAOs
- Constants and interfaces for different scrapers
- Dependency injection and testing
- Type safety (as far as Python supports it)
- Pandas for data handling instead of nested Python dictionaries
- (lots more, but you get the point)
- Integrate official MLB API (which was released after we initially built this)
- Fullstack integration, rather than decoupled backend on AWS & frontend hosted on Webflow
Conclusion
I'm proud of what we built, especially given the experience we had when we started. I still tinker with small improvements here and there, but in general it's nice to look back at this project and see how much I've grown as an engineer.
Projects like this remind me of why I got into software engineering in the first place. Combining my love of sports and math is so fun, and I love that I have the ability to build something unique.
If you're interested in this project, check out Outlier Projections. My friend manages the website and social media while I tinker in the engineering backend.