Let’s take you through a quick rundown of the top 5 things we’ve learned over the years. Our clients have offered us some unique challenges, which have allowed us to grow and establish our reputation as a reliable, ethical, and highly-skilled data services company.
Expect the unexpected
When we were asked to record and annotate toilet flushing sounds, we were a little surprised. It was definitely not an ordinary request. However, when we learned why this health-tech company wanted it, we were totally on board.
The client was developing an app that could keep track of how often a patient used the bathroom as an indicator of possible diabetes.
Knowing that this was a serious health app made us all the more determined to do an exceptional job of capturing every possible variable of toilet flushing sound data.
Read about it in our case study:
Training AI with the Sound of a Flush
Set a high bar and strive to exceed it
An online gaming company came to us because they wanted to train an LLM to weed out and stop the online swearing and abuse within their gaming communities. Establishing guidelines for decency for online behavior is one thing, but going through some of the worst of the worst abuse hurled at other players in a game, and tagging that was quite another. From all that abuse data, we then created scripts to train the LLM, doing our best to make sure we captured all necessary permutations so the custom AI would be able to identify abuse correctly and recommend the right countermeasures predetermined by the company.
Based on some of the truly horrendous things we encountered in the data, we knew we had to get this right.
Read about in our case study:
Behind the Screens: Fighting Hate with Data
If we don’t know, we’ll make sure we learn quickly
A company that makes braces and other devices that help correct misaligned teeth came to us because they were building a database of jawline images to train their AI to show technicians the different types of jaws and teeth out there. The idea was to train the AI to train the technicians. Once we had parameters, we set out to recruit candidates in four continents to ensure we got a large enough representative sampling. The first batch of photos we received back was not usable, so we had to go back to the beginning and establish very detailed directions for those taking the pictures and those posing for the pictures. It was a learning experience for us in terms of how to set up the parameters for data collection for something this specific. We were able to quickly adjust to ensure the success of the project.
Read about it in our case study:
From Angles to Algorithms: A Global Dataset for Dental AI
Cultural sensitivity is essential
Dealing with regional dialects is one of those areas that requires cultural sensitivity because the speakers of those dialects often belong to historically minority groups. Taking on the project of recording regional dialects meant we would have to be understanding of the cultures around these dialects we were recording. We would have to overcome the reticence of the people we wanted to recruit for the recordings.
Find out how the project went in this case study:
Training a Smart Home Assistant to Understand Dialects and Accents
Have a deep database of SMEs and linguists to ramp up a project quickly
Throughout the many different requests from our clients, we’ve noticed one constant, and it’s helped us be successful in all cases. Be prepared, have lists of people to contact to gather data, annotate data, and test the data once it’s in the training model.
When we were asked to compile an annotated database of everyday items we might encounter in our daily lives, we quickly figured out we’d need to cast a wide net to meet the client’s requirements. Fortunately, we were able to quickly sign up the right people, drawing on our database.
Read about it in case study:
From Blurred to Found: Empowering Vision Through AI
Experiences and lessons-learned like those we highlighted here are what make DATAmundi nimble, adaptable and able to rise to most any challenge. We can provide curated datasets, offer annotation and tagging, testing and benchmarking, and translation.
Work with us to meet your GenAI goals.
