It’s a very exciting time to work at Hyperscience. Our New York office is moving into the highest available office space in the western hemisphere — the 88th floor of One World Trade Center. We were also recently recognized as one of Crain’s New York Business’ best places to work, which is a testament to our exemplary culture and dedication to learning.
It’s inspiring to think about how far Hyperscience has come in recent years, says Director of Machine Learning Akhil Lohchab. In fact, when Akhil joined Hyperscience back in 2017, the company was fairly small, with only a handful of people in the NYC office. “I remember being blown away by the handwriting recognition demo we had four years ago,” he says.
Always on the heels of innovation, we’re expanding our handwritten and printed text recognition capabilities to read more languages (like Italian and Arabic!), and improving extraction from unstructured text or tables. We chatted with Akhil to learn more about the exciting challenges the ML team is solving for today, and how he prefers to spend his time outside of work.
What’s your favorite part of your day-to-day life at Hyperscience? What are some example projects the ML team tackles?
My favorite part has been the daily standups with all of the teams that report to me — teams that are focused on semi-structured extraction and Natural Language Processing (NLP). Checking in with my coworkers every morning is something I look forward to, especially since the majority of us are still working remotely due to the pandemic.
The ML team works on a wide variety of problems across computer vision and natural language processing/understanding. We also work in low resource environments. Ultimately, we want our models to improve over time by using as little data (labeled or otherwise) as possible. [Editor’s Note: #ICYMI: Get to know CF, our VP of ML]
What’s the most interesting Engineering project you’ve been involved with during your career to date?
One of the more impactful projects I’ve had the opportunity to work on was the deduplication of documents for an independent federal agency. This agency deals with disability claims, and each claim consists of a highly variable and large number of documents. What made this situation challenging was the absence of a clear definition of what constitutes a duplicate and the scale. I routinely saw documents that were thousands of pages in length. Employees would have to manually compare a 3,000 page document to a 1,500 page document and identify which pages were duplicates of another page they’d already seen. The time it takes to adjudicate on a case is directly proportional to the number of pages one has to go through. Therefore, removing duplicates had a direct influence on how quickly a claim was resolved.
What advice do you have for someone interested in working at Hyperscience or getting started in a career in ML?
The barrier to entry for a career in ML is getting lower every year. There are a lot of tools and libraries out there that make it relatively easy to get started with using — or even training — your first deep ML model.AKHIL LOHCHAB
Examples include Hugging Face, fast.ai, and Keras.
Understanding the fundamentals, however, is still important. Developing a strong ML foundation is essential if you want to turn it into a career. Additionally, being able to write solid code will set you apart from other candidates. At Hyperscience, for example, the role of an ML Engineer is a blend of an ML Research Scientist and Software Engineer. Our ML Engineers are responsible for not only developing new models, but also writing production quality code so those models can be used by our customers.
Do you have a favorite productivity hack?
My current favorite productivity tool is Notion. It’s an all-in-one workspace that connects teams and documents. It’s great for removing silos. I use it to take notes and manage tasks and deadlines.
I also avoid using my phone as an alarm, and put it away in a different room when I go to bed. I’ve found limiting screen time right before sleep and after I wake up extremely helpful for my overall mindset.
What do you enjoy doing outside of Hyperscience?
Although it’s not as easy as it once was, I love to travel and explore new cities. I also enjoy trying new foods and watching movies. I used to play football (known in the U.S. as soccer), and although I don’t play anymore, I still follow it religiously.
Want to learn more about ML and AI but unsure where to start? Check out Hyperscience Learn – an educational program developed by our team aimed at knowledge-sharing and community-building within the ML community.
If you’re interested in solving ML challenges similar to what Akhil discussed above, we’re hiring across the globe. Check out our open positions here.