For some companies and industries, data has become like the air they breathe—it’s all around them, and essential to business growth. For others, data is like a vast and confusing ocean that threatens to drown them in its expansive tides.
Professor of Computer Sciences Jignesh Patel grasps the nuance of both metaphors. Finding a way to help business and individuals harness vast streams of data is the primary reason he and his student Rogers Jefferey Leo John founded DataChat, a tech startup that develops a software platform to make managing data easy for just about everyone.
Patel began researching the idea for DataChat back in 2015, just as the nascent data science field was beginning to explode.
“People were looking to find patterns in data by doing this new form of science, which has a very systematic, methodical way of sifting through the data, building models and applying tools like machine learning to discover interesting trends in the data,” says Patel.
Ironically, there was little interest in Patel’s idea when he first pitched it six years ago. He applied unsuccessfully for multiple research grants on campus, but, undeterred, invested some of his own money instead. Within a few years, both the National Science Foundation (NSF) and the Wisconsin Alumni Research Foundation (WARF) signed on as supporters, allowing Patel to recruit a small number of UW Computer Science students to turn this research idea into an initial prototype.
Things are a little different now. In September, DataChat secured $25 million in Series A funding from multiple venture capital companies to scale up its business model. The field of data science, meanwhile, continues to explode. UW-Madison added Data Science as a major less than a year ago, and more than 500 students have already declared.
Wrapping your head around data science can be tricky. Patel likens it to solving a 1,000-piece jigsaw puzzle. Faced with that challenge, what’s the likely first step in the problem-solving process? Start with the corner piece, of course.
“Then maybe I see a little bit of red in the top corner, and I'm going to try to find an edge piece that has red to try and pattern match and try to get to that next level of solving the problem,” he explains. Humans are very good at recognizing patterns.”
Moving a puzzle piece to carry out the next step in problem solving is easy. But with data, taking that next step typically involves something much more. Years of programming and computer science training is typically required to get to the “how” of organizing and visualizing data. That’s a seriously time-consuming and expensive approach.
“Why do we require someone to take many years’ worth of training in programming?” asks Patel. “Why can’t we just keep them at the level of thinking about the next step?”
DataChat accomplishes that, through something Patel calls “conversational intelligence.”
Most of us are familiar with chat bots—those applications that let us navigate basic Q&A interactions with insurance companies and stores online. But even though the word “chat’ is in its title, DataChat is focused on fostering a conversation about data between the user and the platform. The user asks questions and issues commands using a specific, controlled language—DataChat English—directing the platform to accomplish a wide range of tasks, like visualizing sales data or using machine learning to find root-caused for underperforming students.
Having that controlled language is key to the platform’s functionality. Introducing even the slightest amount of ambiguity into the conversation can be disastrous, says Patel.
He goes back to his puzzle metaphor.
“Think about building a 3D puzzle,” he says. “If you put the wrong piece in the bottom layer, that's going to cause you trouble later on, so we’ve set things up so that you can easily recognize that pattern and easily correct it.”
Patel has structured DataChat to cater to different types of users—the casual user who just wants to look at high-level trends and the more serious domain expert who wants to get into much deeper detail. As users become more familiar with the platform’s conversational language, they can leverage their own expertise to learn more, like a retailer looking at product pricing data or an economist parsing financial reports.
“Who doesn’t have an Excel file? Who hasn’t taken a course where they have an assignment that had something to do with data?” Patel asks. “We cater to the entire spectrum because sometimes the journey just starts with something as simple as just visualizing the data.”
Another key piece of DataChat’s appeal is its reproducibility. The platform remembers the conversations, and they become like a recipe that makes future queries simple to execute, even when it involves new types of data.
Funding firmly in hand, Patel and his staff are currently focused on the particulars of scaling up—DataChat has grown from the initial team of three students in 2017 to a staff of 40 full and part-time employees, located in offices on campus and in Research Park on Madison’s West Side. Many of those employees are UW students who have graduated with Computer Science and Statistics degrees. Given the skyrocketing interest in data and data science, the future’s looking particularly bright.
“There's a need for our platform everywhere,” says Patel. “It just needs the customer to be innovative and say I'm going to build a data science strategy differently—by empowering every data user to take on a data science journey, rather than just relating that power to data programmers.”