Applied machine learning
Personalization, ranking, and on-device inference, with an evaluation harness before anything reaches a user.
What this engagement is
We apply machine learning where it changes the product: personalization, progression, ranking, and classification. The deliverable includes the evaluation, not only the model file.
If a rules system would be clearer and safer, we will say so. A model is a choice, and it has to earn its place.
How we build it
We define the failure that matters, then build a set of cases that can catch it. Outputs are checked before release. For health-adjacent work, that check is against published evidence and a human review, not a vibe test.
On-device models are preferred when the data is personal. Cloud inference is used when the task cannot run locally, and the boundary is documented.