Amid the hype surrounding AI, it’s easy to forget how important people are in the process. For our purposes, AI requires human intervention to train a model to extract the correct data from the documents and also to perform quality checks: a process known as annotation. The human element in AI systems creates what is known as a ‘human in the loop’ or a ‘HITL’ system and ensures that the AI model is fit for customer use.
Checking a document is a tiresome process that requires an exhaustingly high level of detail. Evolution AI’s commitment to accuracy requires our team of annotators to work hard to pick out minor errors and train our AI model not to repeat them. And, as Evolution AI processes over 1.5 million documents daily, our annotators face a formidable workload.
Recently, an issue with our User Interface (UI) kept appearing, considerably slowing down our annotation rate. Our old interface was slightly confusing, meaning that sometimes batches of documents got mislabelled during the customer uploading stage and consequently incorrectly processed. The result was a giant tangle of data that brings a dismal look to the faces of our annotators whenever it is mentioned.
From there, we began to investigate. More and more issues began to come to light: our UI layout was making it hard to compare the table of extraction data to the uploaded documents, and so forth. One or two of our customers had also noted that our interface seemed outdated compared to the quality of the AI technology behind it.
Clearly, our old UI was holding us back and, to a large extent, wasting our time. Therefore, we made the decision to embark on an intensive project to update our UI to become sleek, modern, and optimally user-friendly. Throughout the following year, nearly every member of our team became involved in the journey to design and implement a new UI that would impress our customers and also protect the sanity of our annotators.
Working with a designer brought its own set of challenges. For starters, aligning all of our staff’s visions with that of the designer required a large amount of back and forth. Alongside this process, we carefully pared down the old interface to its simplest elements while maintaining functionality. And something was emerging: a new interface that preserved the robustness of its predecessor but had the intuitiveness of a modern app.
Of course, the project succumbed to some of the pitfalls of any large project: it ran over schedule. At times, it seemed like it was stagnating. But we got there and developed something that we’re sincerely proud of.
In our brand new UI, there are several automation features that we’ve made prominent. Our new dashboard (see below) offers users a comprehensive look at the data extraction processes. We’ve also introduced a labelling system to increase transparency during the extraction process. Another small but nifty feature of the new UI is that our customers can now tag their documents with arbitrary information. This metadata tagging feature allows our customers to easily organise batches of extracted data, reducing their post-processing time.
And, perhaps most importantly, it’s fast. The intuitiveness of the new UI model means that our annotators can breeze through batches of documents instead of being bogged down by minor errors.
Anyone is welcome to try out the new UI; simply book a demo.