June 17, 2021
Hadayat Seddiqi, Director of Machine Learning

InCloudCounsel Spotlight: Hadayat Seddiqi, Director of Machine Learning

There was no cable television in Hadayat Seddiqi’s home when he was young. So instead, Seddiqi, InCloudCounsel’s Director of Machine Learning, spent his days trying and “miserably failing,” he quipped, to decipher his father’s computer programming books.

Things clicked when he was nine and began tinkering with websites. In high school, he learned to code, but made a hard pivot after taking his first physics class. He wanted to be a scientist.

From there, he went on to study astrophysics, colliding galaxies, and quantum physics. After college, he spent a year with NASA as a research associate in the Ames Research Center near San Francisco working on artificial intelligence systems for robots destined for space. Next came the Oak Ridge National Laboratory where he studied quantum computing, the union of his two passions: computer science and physics, and published a handful of academic papers. Again he pivoted, shifting from Oak Ridge to algorithms and software engineering at biotech and consumer analytics firms.

Seddiqi joined InCloudCounsel in 2017 and was tasked with building its machine learning and AI programs to analyze and understand legal contracts to augment and automate legal services. Since then, the company, its customers, and Seddiqi’s team has grown on its mission to free professionals from high volume, repetitive contracts so they can focus on their most important and most exciting work. 

How do all of your academic and professional experiences play into the work you do today?

There’s a common thread of leveraging applied math and computing power to do science. I’ve always loved learning how to translate the information about the world into a language a computer understands, and push the limits of our technology to help us do amazing things.

When I worked at the quantum computing lab we had what was once the largest supercomputer in the world. Seeing it work was remarkable. It was like opening a door to something way, way bigger than anything I ever thought possible, and it exposed me to the enormous challenges of massive scale computing. On the algorithms side of computing,  I remember building an AI model that could recognize handwritten digits and that was really eye opening. These tools are not only useful for predicting the future, but are smart and can quickly learn to identify totally new patterns.

I have a constant urge to reach into other disciplines and my colleagues at the quantum computing lab were the same. That made it easy to do pioneering work in quantum computing because most people weren’t exploring AI at the time. I’m very cross-disciplinary by nature, and I try to bring that spirit into my work today as we build toward computable contracts.

Tell us a bit about the history of machine learning at InCloudCounsel. 

Before I joined, our CTO Lane Lillquist and Senior Engineer Rich Niles asked a question: Is it possible to take a legal document, ask a question, and build an algorithm to predict the answer? The question was whether there was a non-solicit clause in a contract. The answer was yes. They quickly decided they needed a machine learning function. 

In the beginning, it was just me. All we had was lawyers working in our system processing routine contracts and we were tracking the data. We had an algorithm that would comb contracts looking for the relevant clauses and predict their appearance, but it wasn’t accurate. We realized we needed to start highlighting and labeling documents. We became a team of two, and began to make more headway. 

It took about a year to reach another plateau. There, we realized many of the questions we wanted to answer were created by lawyers who baked all sorts of knowledge, logic, and assumptions into them. What we needed was to reframe the questions to work in a machine learning system while taking into account our customers’ and lawyer partners’ needs. We needed a common language. Just getting to this point was difficult. It required multiple rounds of failed iterations trying to understand whether our algorithms were correct, and if we had the right data sets for them to interpret. We’re very fortunate because as a company we’ve processed nearly 500,000 contracts. We have the data. One of the most important parts of our job, and one of our biggest accomplishments, has been to ask the right questions, collect the right data, cut through the noise, and analyze it in a way that yields increasingly accurate contract analyses. 

What is machine learning focused on now?

Today, we’re a team of four. By the end of the year we’ll be seven. Recently, our wider team, which includes engineering and product, finished refining our knowledge graph which describes a contract in a way that is understandable for a machine learning system, and functional for the business side of our company and our customers. 

This is a really fundamental, powerful tool because it helps us understand, not just the contracts themselves, but their analyses for both our customers and our future as we seek to build better, more intuitive systems. It helps us put structure to it. Additionally, now that we’ve created a graph for one specific contract, we can create them for others far more quickly and efficiently. It’s a modular system. Jurisdiction, for example, is a provision that’s in most contracts. We’ve already built for it. Going forward, we’ll be able to repurpose it to start running machine learning programs faster and faster. 

What are the short and long term goals of machine learning at InCloudCounsel?

This knowledge graph we’re creating is a higher level language that ties together the important data from our document processing system and soon from different parts of the company. It’s our Rosetta Stone. It allows us to “speak” in a way that’s consistent with our various functions and our machine learning applications, ultimately giving us and our customers better insight into routine, high volume contracts and ways to streamline them. 

The volume and kinds of documents we’re processing is continuing to grow, and that’s to our benefit. There’s a virtuous cycle where we constantly feed data into our models, and the more that goes in the more accurate our results become. We’re able to incorporate the knowledge and intuition of lawyers in a system and refine it. At the same time, we’re building a system encoded with rules to seek out a specific term or phrases, labeling the results, and running that through our machine learning models. These models are really good at finding underlying patterns, and delivering results that may appear incorrect, even to an educated analyst, but turn out to be right. This efficiency gain is critical given the hundreds of concepts we need to model.

Many organizations are plagued by what’s called the cold start problem, in which there isn’t enough data to conduct analyses, delaying or precluding their machine learning initiates. Luckily, we don’t have this problem. The more data we gather on preferred contract terms and underlying negotiation trends the more accurate our predictions will become, bringing us closer, one day, to truly computable contracts. 

Learn more about how InCloudCounsel’s purpose built technologies help free enterprises to focus on their most important work and bolster their business relationships

Hadayat is the Director of Machine Learning at InCloudCounsel. Prior to InCloudCounsel, Hadayat worked on space exploration robots at NASA, published papers in quantum computing at the United States Department of Energy, and engineered DNA sequencing software at a biotech startup. Hadayat holds a B.S. in Physics from Georgia Southern University. Outside of work, Hadayat enjoys nature, rock climbing, and snowboarding. Connect with him on LinkedIn.