Building an ML-driven system is a challenging feat, especially when it involves collecting quality datasets. In our case, the goal was to gather multiple pictures of the same horse to fuel our Equid Biometric Verification system. The task was indeed daunting but offered us an opportunity to tap into the power of a data flywheel.
Our design approach was to create a system that immediately adds value for users, even while the ML model continues to improve with incoming data. In other words, each user interaction with the system simultaneously enhances the user experience and nurtures our ML models.
For example, to facilitate event check-ins, our application uses the Equid Biometric Verification system. This process not only streamlines the check-in procedure but also provides additional longitudinal data points. Every new picture taken for check-in becomes a valuable data point that strengthens the system's performance over time.
Crucially, we've also incorporated a human override feature. This means that even while our AI is improving and learning, human insight and experience have the final say, thus ensuring that the system is robust, reliable, and accurate.
As users continuously interact with the app, the ML component becomes increasingly effective and accurate. This creates a data flywheel, where the user experience and ML model mutually reinforce each other, leading to an increasingly better product.