I wanted to dumb it down a bit, leave the fancy equations out of this blog post, and talk about what a Software / DevOps / AI Engineer could do for your NASCAR team. I’m inspired by an application question I filled out for Hendrick Motorsports that asked a similar question, and I don’t believe a 300-character-limited space did my response any justice. Let’s begin.  

Data Science

Every morning I make my omelet and 1 ½ cups of coffee followed by sitting down and checking my emails. Every single morning unless I have to wake up with the roosters and drive a family member to the airport. I am predictable. So are NASCAR drivers. My job is to let you know how many cups of coffee they drink everyday and whether they like grape or orange jelly on their toast.

Opinion Warning! I find that much of the NASCAR data collected focuses on the car’s performance and is not necessarily a reflection of the driver. I’ve changed out brakes and done a few oil changes in my lifetime, but I’m not a master mechanic. I’m excited to maximize the performance of the car as well, but I just think the focus of my job should be collecting data geared toward driver performance and pit crew optimization. Yes, the pit crew—my fellow jocks. Does Tommy B. lose a second by stutter-stepping with his left foot instead of his right? The mechanics of the pit crew don’t differentiate much from a sprinter’s starting stance or a high/long jumper’s initial steps. Long step or short step? Does the pit crew perform better after a night of drinking, or are they fumbling around the car?  For the driver I’d ask, does Dave the Driver do better overtaking a  car from the right or left. This strategy, that strategy. What works. What doesn’t, optimal drafting times during green and caution flags. Other driver data is just as important. Does #22 close the inside as you approach from the left every single time allowing you to quickly pass on the right? I want to make an average driver great or a great driver greater. That’s why you should hire me.

Software  

I can build enterprise-level software, and I can build it fast. Lots of fancy buttons collecting data from simulators—and not just the usual data. Non-traditional derivatives calculated from lane lines, drafting time, drafting position, and more statistics, with passing strategies as a reflection of the right side of a negatively skewed bell curve. Pit stops and prolonged drafting on the left. In summary, I’ll bloat the living shit out of the software until one of you tells me, “Take that shit out of there. We don’t need it.” If Joe Gibbs called me at 5:00 AM and said, “Pete, I need an application that does this, this, and that, and I need it in my inbox by 5:00 PM end-of-day,” I’d say, “Sure thing, Sir. Please send any details you’d like about display, buttons, or functions throughout the day.”  

HPC / DevOps

I know AWS, GCP, and Azure (which mirrors AWS). Which one do you want to delve into?

• Public Subnet: For AWS, I can set up an XL EC2 (Elastic Compute Cloud) VPC (Virtual Private Cloud) environment with public and private subnets protected by NACLs for stateless traffic from our team’s IPs only. Our network load balancer would ingest UDP data to the routing table, sending it to the public subnet.
• Private Subnet: AWS Batch to run Slurm to manage the cluster queues and resources, along with an Elastic Fabric Adapter (EFA). Kinesis would process the data decoupled from ingestion rates and send the data to an S3 endpoint via a VPC gateway.
• S3 Storage: Even billionaires don’t want to pay an arm and a leg for storage, so I’d write a quick Bash script to automate Glacier archiving, or I’d use the AWS Console’s Lifecycle Manager.

AI / ML  

You’re in control. I can figure out what I think we should know, but ultimately the team leaders are the experts. Has the Crew Chief ever had to “guess” when and what type of pit stop to make? I can make that guess an advantageous decision that favors a positive outcome by 50%, 60%, or even 70+%. In short, if the Crew Chief tells me where they’re struggling, I’ll quietly run some calculations and turn those uncertainties into their strengths.  

The same goes for drivers. Do you have better luck passing #11 on the inside or outside? Just like baseball batters, drivers have their strengths and weaknesses. I’ll have a profile on every last one of them. The algorithms don’t have to be overly complicated. We’re trying to find trends of winning race teams for every course. The winning team becomes the benchmark for our data, whether you like them or not. My Python scripts would explore Scikit-Learn, XGBoost, and PyTorch. For small pit stop datasets, I would use Scikit-Learn for lap time prediction based on tire degradation rates. Also, is it ideal to cut out a 12-second pit stop entirely? 4 vs. 6 total pit stops? Is the benefit of a standard 4-tire change pit stop worth it 3, 4, or 5 times at the loss of 10–12 seconds per stop?  

Data Collection  

We’re going to need interns to watch races and collect specific data from a bird’s-eye view of the race. From what I understand, NASCAR doesn’t have footage of the entire race from a stationary point from above. I would have an intern film the entire race from the top row of the stands with a camera mounted to a tripod. This is probably the single most important piece of data gathering for driver strategy. It seems there’s an obsession with micro-data and derivatives, and not enough focus on macro-data zoomed out to see the big picture.  

In conclusion, that’s what I’d do for your NASCAR team if you hired me. Having coffee on-site and ordering the occasional NY style pizza or deli sandwich while I’m working is greatly appreciated.