Rogers Jeffrey Leo John, Ph.D. co-founded DataChat in 2018 and today serves as Chief Technology Officer. In this interview, Rogers talks about the origins of DataChat and how it has transformed the workflows, audience, and impact of data analytics.
How did DataChat come to be?
The concept of DataChat formed in 2016 after Jignesh Patel, my co-founder, served as a visiting scientist at Pivotal Software, a cloud hosting and consulting company. He noticed that at Pivotal, the process of solving data problems followed a pattern. The data team always ended up in a loop of loading the Python package, selecting features, training the model, then analyzing the results. This was followed by tweaking the features and retraining the model over again. Repetitive workflows like that are ripe for automation.
With that observation in mind, Jignesh and I wrote our first paper, which suggested that a conversational intelligence assistant could abstract the model training loop into a Python template. In other words, we proposed that a user could solve data science problems by writing directions in English. We spun our research out into DataChat in 2018.
How exactly did you improve the model training process?
We made it possible to train models without any coding skills. To do so, we leveraged controlled natural language (CNL) to abstract away programming languages like Python, R, SQL, etc. DataChat’s CNL is called Guided English Language©, or GEL.
DataChat users never actually see GEL. Think of it as an aviation “language” like the NATO phonetic alphabet, which is probably familiar from movies like Top Gun. The words Alpha, Bravo, Charlie, Delta, and Echo translate into the letters A, B, C, D, and E, right? GEL is like that, only it translates a user’s words into SQL and Python.
How has DataChat evolved over the last six years?
We’ve taken DataChat to the frontiers of machine learning and explainable artificial intelligence (AI). Our AI not only translates natural language into code but also performs complex data science functions in response to prompts written in plain English. The people who know an enterprise best can ask questions about complex data regardless of their technical abilities.
Talk about reproducibility and why it’s a priority for DataChat:
A few years ago, enterprise data teams didn’t care about reproducibility, which is the principle that scientists should share their conclusions along with the process that led to those conclusions. Reproducibility is the difference between making pretty graphs and doing true data science.
So, we baked reproducibility into DataChat. When users converse with DataChat, the platform automatically documents the conversation as steps in a data science process. That way, others can validate the workflow or reapply it to new data. The documentation is in English, which is useful for data governance and transparency.
In your view, who is DataChat for?
DataChat’s vision is to make data meaningful and accessible. Every organization has untapped data, but few employees besides data scientists have the tools and skills to make use of it. DataChat solves that problem. We think the people who know the business best, regardless of their department or technical background, should have the tools to question data on their own terms. We believe the most successful companies will be those in which everyone can make data-driven decisions confidently. DataChat can make that happen.
What’s next for DataChat?
Since launching, DataChat has made a significant impact for enterprises, including those with practically unlimited resources. The fact that companies like Meta use DataChat to solve expensive business problems tells us that we’re addressing a gap in the analytics market. We’ve spent many years honing our technology. Now, our priority is to bring DataChat into companies with underutilized data, which represents untapped potential and value.