Projects
What we have built so far, how it works, and where it falls short.
Results
How we work
We test on data the model has never seen
We hold back 900 of our 3,000 transformers until the very end and check them once. That is the only honest way to know if a model works.
We say when we are unsure
Every prediction comes with a confidence score or a range. When confidence is low, the answer goes to a person to review.
We can explain every line
Each of us can walk through the code and defend the choice behind it. If we cannot explain it, we do not use it.
More projects will appear here as we finish them.