Machine learning is the holy grail of data analysis, but unfortunately, that holy grail oftentimes requires a PhD in Computer Science just to get started. Despite the incredible attention that machine learning and artificial intelligence get from the press, the reality is that there is a massive gap between the needs of companies to solve business challenges and the availability of talent for building incisive models.
YC-backed Intersect Labs is looking to solve that gap by making machine learning much more widely accessible to the business analyst community. Through its platform, which is being launched fully publicly, business analysts can upload their data, and Intersect will automatically identify the right machine learning models to apply to the dataset and optimize the parameters of those models.
The company was founded by Ankit Gordhandas and Aaron Fried in August of last year. In his previous job, Gordhandas deployed machine learning models to customers and started working on a tool that would speed up his work. “I actually realized I could build a version of the tool that was a little more advanced,” he said, and that work ultimately led to the foundation of Intersect Labs. He linked up with Fried in October, and the two have been working on the platform since.
Intersect’s goal is to move analysts from purely retrospective analysis to creating models that can predictively determine business strategy. “People who live in SQL and Excel, they are really good at pulling the data of the past, but we are giving them the superpower of seeing the future,” Gordhandas explained. “All you need is your historical data, upload to our platform, and answer two questions.”
Ankit Gordhandas and Aaron Fried of Intersect Labs. Courtesy of Intersect Labs.
Those questions essentially ask what the model should predict (the outcome variable). From there, Intersect begins by cleaning up the data and ensuring that the various columns are properly scaled for data analysis. Then, the platform begins constructing a range of machine learning models and evaluating their performance against the target output. Once an ideal model is identified, customers can integrate it into their other systems through a REST-style API.
What’s interesting here is that Intersect can get better and better at identifying models over time based on the increasing diversity of datasets that it gets access to. Plus, as researchers identify new models or ways to tune them, the platform can potentially proactively improve the models it had previously identified for its customers, ensuring that they stay at the cutting edge of the field.
Today, the platform can handle one table of standard rows and columns for processing. Gordhandas said that the company intends to expand in the future to “image processing, audio processing, video processing, unstructured data processing” so that the platform can be applied to as diverse a set of data sources as possible
Gordhandas says that Intersect is attempting to sit in the middle of more specialized machine learning platforms that are limited to hyper-focused niches, while also offering more analytical power than comparably simpler solutions.
Certainly the space has seen a proliferation of options. New York City-based Generable (formerly Stan) uses Bayesian modeling and probabilistic programming to improve drug discovery, while Mintigo uses AI modeling to improve customer engagement. A huge number of other startups target different stages of the data analysis pipeline as well.
In the end, Intersect hopes to make these tools more widely accessible. The company has a couple of early customers already, and is going through the Y Combinator accelerator this batch.