Model Development

Once we've analyzed a dataset and found signal, we proceed to actually build, evaluate and improve models. We use an agile iterative approach and rapidly build, assess efficacy, and release suboptimal models. We do this intentionally so we can focus on de-risking your business hypothesis as quickly as possible.

If you, for example, have a hypothesis that AI can be used to identify a burglar, we might quickly build a person-on-the-premises model. Knowing if there is a person on the premises, is a pre-requisite to next determining whether that person is a burglar. We would push to get this model fully accessible through a secure easy to integrate API so your product folks can start to play around with the idea in a UI of your choice. Next we might hone in on examples where actual burglars (in this case presumably actors pretending to be burglars as we're still jumpstarting the project) can be detected with some level of efficacy. Again, we'd release this cruder model earlier, to start teasing out how to deal with the inevitable errors from a product vantage.

Once the business / ML model fit is established, it is seamless for us to transition to a production status. Presuming you want and will benefit from more effective models, our machine learning experts will then, behind the scenes, continue to improve the model efficacy. Our experts will then seamlessly release new models to production as they are ready.