Edgett Hilimire is a developer at MMIT. He has worked professionally as a developer since he was 15, and joined the company after working on implementing an artificial intelligence (AI) system for an asbestos claims company. He currently works on developing new software that uses artificial intelligence to improve MMIT’s document collection processes.
Q: Tell us a little bit more about your role at MMIT.
A: I work on developing software that plays a critical part in our collections process. We use the Elastic Search database, which is a new technology that we use to house all of our documents after we collect them. The Elastic Search database then feeds documents into MMIT Reach for our clients to access and into our internal PAR (policies and restrictions) “hunting tool” that helps us surface new policy and restrictions changes.
Q: What’s your favorite part of your job?
A: One of the great things about working here is you’re free to use new technology as it pops up and as it is useful. As a team, we are always looking for ways to incorporate new technology to help resolve any business needs. Everyone will branch out into new technology zones and build proof concepts that they share with the group, which anyone can use to resolve any future problems. We’re able to continuously move forward on our usage of new technology because everyone participates in it. It’s all about how we can use new technology to come up with better solutions.
Q: Is there any new technology that clients should keep an eye on?
A: MMIT has been pushing implementing and improving our use of AI. If anyone is interested in using AI, my advice is to start now. It’s a valuable resource if you take the time to fine tune and train it. The biggest part of implementing any AI system is the feedback loop where humans validate the data coming out of the AI and then feed corrections back into the system. Then AI takes the corrections from this feedback loop and will apply it going forward, and in turn it becomes more and more accurate.
The feedback loop is the trickiest part of the AI, because if you’re not feeding back data into the AI, you’re not doing it right. Once the humans have given the computer enough information to do its job, the computer will always be faster.
Q: Can you go into detail about how MMIT uses machine learning and AI in our tools?
A: Right now, we’re focusing our AI on classifying documents. When Elastic Search feeds a new document into the PAR hunting tool, the AI automatically tries to classify the document as a policy, form, etc. We’re working on implementing a screen in the PAR hunting tool where the AI will classify the document, and then the hunting team will have the option to correct the AI if it tags the document incorrectly. The AI will then take those corrections and will apply it on its next round of classifications. The more the PAR team corrects the AI, the more accurate it will become in classifying documents.
We’re also using the AI to automatically check if a document is an updated version of a previous document. The AI will look at the two documents to find differences in wording to determine if the document is an updated version of a previous doc.
Q: What’s been your biggest victory with the company so far?
A: I would say the collections system has been my biggest victory. The AI can go out and collect new documents every day, and store it in the Elastic Search database. Then the documents feed out into our different tools. It was real team effort to get this collections system to work, and it was an immense success.
I’m not ready to call victory on the AI, but I can smell victory. It’s difficult to work with, and has required input from the development team to the PAR hunting team, so again it’s a real team effort. We all have to understand the end goal and be willing to suffer through a lot of difficulties to produce a system that works and works well.
Q: What are some of the challenges of your role?
A: The challenge is the speed at which we move. Sometimes we move too fast and we pump stuff out and there are times when it doesn’t work, but to me this isn’t always a bad thing. We’re able to learn from our mistakes and improve from them. To me, it’s better to fail early then to fail late. Managing the business desire to get things done now and getting things that work properly can be a challenge.
Q: What do you like to do outside of work?
A: I like to sail, and I like to travel. I’ve traveled to all 50 states except for the northwest region. I’ve also been to Canada, Mexico, Scotland, Guatemala and the Philippines, which I really enjoyed visiting. I might go to Portugal this summer, and I think I’m going to visit California to see some of my friends that live out there.