Modern machine learning methods have a lot of power, and many limitations and we want to build models that work well and that users can understand. We know that sometimes there's a trade-off between those two and we'll design a product both with you and for you.
Some Past Projects
- Designed a neural network that determines data file type being submitted by clients without needing to parse the file so it can be inserted into a database or flagged for possible errors for an Ed-Tech non-profit.
- Utilized an association rule learning model to help researchers determine common K-12 course trajectories, and polynomial logistic model to identify students who could succeed if given better guidance, for an education research group.
- Designed a bootstrap-aggregation ("bagging") algorithm to create matched samples, when propensity-score and other modern matching methods wouldn't suffice, for early childhood development researchers.