Football
Analytics
AI

Raw football datasets contain hundreds of variables — most of which add noise rather than signal. The challenge was identifying which features actually predicted outcomes and building a model that was accurate, not just large.

A secondary challenge was development speed: building, testing, and iterating a machine learning pipeline from scratch with a small team is slow. Finding ways to accelerate without sacrificing quality was equally important.

Two parallel strategies drove the results:

• Data preprocessing & feature selection: Systematically reduced the feature space by identifying which variables had meaningful predictive power ( Touchdowns, yardage gained, quarter) — removing noise that was degrading model accuracy
• LLM prompt engineering: Used large language models as a code generation accelerator and guide — writing precise prompts to generate, debug, and refactor pipeline components instead of coding everything from scratch

• Data ingestion and cleaning pipeline to handle raw football statistics
• Feature selection analysis to narrow high-dimensional data to effective predictors
• Predictive model training and validation with accuracy benchmarking
• Prompt engineering workflow for LLM-assisted code generation and debugging

The combination of better features and faster iteration cycles compounded — cleaner data made the model more accurate, and faster development meant more iterations in the same time budget.

The same skills used here — data preprocessing, feature selection, signal-vs-noise thinking — are core to data acquisition systems in aerospace and energy applications. Sensor data from high pressure systems or flight test hardware has the same fundamental problem: large, noisy datasets where identifying meaningful signal is the hard part.

LLM-accelerated development is rapidly becoming a baseline skill in engineering. The ability to write precise prompts that generate usable code reduces iteration time in any technical discipline.