An AI analysis framework delivered across all 1,586 Arizona K-8 schools: correlation and outlier detection, evidence-first validation against official sources, and per-school outcome forecasts, all runnable on a self-hostable stack.

The challenge

The Helios Decision Center works with rich public data on 1,586 Arizona K-8 schools, spanning academic results, teacher staffing, school characteristics, and community context. The hard part is using it: finding the non-obvious patterns by eye, confirming each number against the official state source, and forecasting where a school is heading is not feasible to do by hand across 1,586 schools and 61 metrics. Decisions about Arizona schools ride on that data, so whether it can be trusted and whether the signals in it are real matter as much as the analysis itself.

What we built

• A correlation and outlier analysis that compares any two of 61 school metrics across all 1,586 schools and flags the statistically unusual ones, in an interactive dashboard
• An evidence-first validation system that checks each field against its official public source (AZ School Report Cards, the Auditor General, the State Board of Education, the US Census) and logs every attempt for audit
• An exploratory predictive layer that forecasts per-school outcomes in math, ELA, and science, surfaces the main drivers behind each, and flags schools performing well above or below their prediction
• A portable, self-hostable stack (database, API, and web UI) so the Helios team can run the whole analysis on its own

The results

• Delivered all three analysis phases across the full 1,586-school dataset: correlation and outlier detection, data validation, and prediction
• Validation traces every field back to its official source and keeps an auditable record of each result, so any number on the dashboard can be checked
• Roughly five per-school forecasts, each paired with the factors that drove it and framed as decision support rather than a black box
• Packaged as a self-hostable stack the team runs independently, with every public data source documented end to end

Built with