From Text to Skills: Building an AI Engine that Understands Reading (2023)

CHALLENGE

An education technology company needed to automatically identify over 200 specific reading skills in any text and generate targeted practice materials, but had limited training data and required results domain experts could validate.

SOLUTION

We created a human-in-the-loop machine learning system combining linguistic analysis, custom ML models tracked in MLflow, and an expert annotation interface that continuously improved through feedback—achieving accuracy gains from 29% to 95%.

RESULT

The system now powers personalized learning for 2,000+ students across 42 schools, with projected growth to 10,000 users within a year, while dramatically reducing the workload for teachers creating targeted instruction.