Jignesh Patel, a University of Wisconsin-Madison computer scientist, wants to make understanding data like turning on lights in a house.
You shouldn't need an electrical engineer on hand for a bulb to illuminate after flipping a switch. Nor should the manager of a business need a coder on hand to get answers to simple data-oriented questions.
That’s why Patel and Rogers Jeffrey Leo John, a recent UW-Madison computer science graduate student, founded DataChat. The 18-month-old startup promises to let decision makers without technical expertise have a “conversation” with data.
“The end user who needs to make decisions with data, if they want to interact with the data directly, they don’t have a means to do that (right now),” said Patel.
Patel, who has long been seen as a leader in the realm of big data, is no stranger to entrepreneurship: He has co-founded three companies, each of which were acquired. Twitter bought one of his enterprises — Locomatix, a data analysis tool for mobile devices — in 2013.
This latest company is a solution to what Patel described as a “last mile of delivery” problem for companies that rely on data. A business may have resources for dealing with the rest of the data "supply chain" — it can collect data, store it in a warehouse and organize it. But when a higher-up actually wants to use that data to help them with decision-making, it can be a complicated and time-consuming process.
That’s a problem, given the role data plays in institutional decision-making in 2018, said Patel.
“Data has become the ubiquitous currency on which businesses live,” he said.
What DataChat offers to solve the problem is a smart assistant, something like an Alexa or Siri of data analysis. The idea is for managers to sit down and type out requests for various data-related insights with a chatbot.
“If you’re a manufacturer, and you’re running the factory floor … they can say, ‘Hey, show me the facilities with the most down time,’” said Patel.
DataChat parses that request, and translates it into a computer query to retrieve the relevant information — a complex form of computing that falls into the realm of natural language processing, a subfield of artificial intelligence concerning how computers understand human writing or speech.
Patel said that to get a computer to answer a plain speak question about data has typically involved programmers writing code to bridge the gap: “They’ve engineered languages to make sure that whatever the human intended comes back with a precise answer,” he said.
The core technology behind DataChat automates that, analyzing a sentence and then inferring intent, something that Patel said UW-Madison computer scientists have been working on for years.
Because of that natural language processing technology, DataChat can spit out a table or a visualization in real time telling a manufacturer which of its sites has the most downtime, and through other sophisticated algorithms, follow up with suggestions for other information it could retrieve.
The technology also incorporates machine learning tools — the ability for a computer to form habits from previous computations. That means DataChat can display or suggest data insights based on what a user looked up previously.
“As you interact with the platform over time, it learns what the user is focusing on,” said Patel. “It knows what you’ve done in the past.”
The company itself has seen some initial success with its product. It has a small batch of customers, and has taken on a staff of 11 employees.
Patel said that DataChat is in the process of acquiring more clients. That said, the company is taking a slow-growth model — the goal is to work closely with a small group of initial clients before expanding.