Learn how Watchful customers are accelerating AI, from production to deployment, to allow their NLP and LLM initiatives to be more cost-effective and scalable.
A Proper High, as with many e-commerce data platforms, has to go through the painful and often manual process of cleaning highly fragmented and noisy product data from thousands of different sources. They budgeted 90 days of development time In order to solve just one of several classification tasks.
In one afternoon, one engineer was able to use Watchful to classify their entire 200k+ product library with greater than 99% accuracy in less than 4 hours of effort. The original estimate of work was approximately 30 days of effort. Watchful has enabled A Proper High to, profitably, deliver a cannabis data solution in a fraction of the time it would have taken with legacy SME labeling.
Investment and trading professionals rely on quality data and insights to drive critical decision making. However, the cost to develop models in the financial domain has been enormous due to the time requirement for Subject-matter-expert labeling. Investors’ time is too valuable to waste labeling data to train models, especially without clear ROI.
The investment firm’s data science team uses Watchful to rapidly and scalably deliver labeled data, by themselves, for several high-impact classification models without having to waste precious SME time. Machine Learning practitioners are able to rapidly prototype, iterate on, and deliver production-ready models without a labeling service and/or team by programmatically labeling the data themselves.
The enormous volume of CRUD events combined with onerous data privacy requirements make getting the necessary amount of labeled data a difficult and time-intensive task with hand labeling. It was projected to take approximately 1 FTE Data Science 1 quarter of effort to achieve reasonable results.
After defining the project and class space, a single data scientist was able to deliver their first production-ready model in less than one day by using Watchful. This approach was then able to be applied to 2 more models which were delivered in one person-day a piece.