Deep NLP with Aerin Kim
Interview with Deep Learning startup BYOR Founder Aerin Kim
Aerin Kim, Data Scientist and Founder of resumé checker BYOR (Build Your Own Resume) uses Phrase2Vec NLP parsing technology to help users improve their CV by examining words and phrases and then, using the Deep Learning parser, suggesting how to make it better.
Aerin will explain some Deep Learning NLP essentials at next weekend’s AI With The Best online conference, a follow-up of her previous talk on Phrase2Vec which you can catch here. We were pleased to have asked Aerin lots of questions last time and happy to see the amazing progress for her startup! You can find out more during her live talk this weekend — but for now, here are her answers to our burning questions.
Q Congratulations on the growth of BYOR labs — what have you been up to since last September?
Thank you! Lots of things. NLP/ML wise, upgrading and testing the phrase suggestion algorithm and the resume bot is something that we do regularly. Engineering wise, we are building a robust testing suite for the bot.
Besides Engineering, recently I became an organizer of Meetup group “Ladies that UX” of 1000 female UX professionals in Seattle. Last year we laser-focused on building the product in a technology driven way but this year we are focusing on how to create a Wow! user experience.
Marketing wise, we’ve been featured on ProductHunt (Yay!) and gave a tech talk at University of Southern California.
Q How have people been reacting to the BYOR AI’s resumé fixing suggestions?
People who have used BYOR absolutely loved it and about 30% of our users upload their resume more than once after they fix their resume following BYOR’s suggestions. It’s difficult for anyone to know how to write an effective resume unless you show it to someone who has already done it before successfully, and get some feedback from them. I think it’s people’s natural tendency to seek feedback and assurance on what they wrote. 70% of our users come from Facebook. We never advertised on Facebook but our users have shared BYOR on their student groups and tag their friends who they think need the help.
BYOR is a 100% machine generated (automated) response and no human supervises the results before they are returned to users. So sometimes its suggestions are not usable in the exact sense but our users have been understanding and encouraging, telling us it still gives them a good idea of how to write good resumé sentences. Maybe it’s because BYOR is a free tool and only the engineering community uses it. But I think it’s also because they like our initiative to make this process automated and are impressed by it.
Q Your deep learning resumé parser has built over 11,000 machine learning datasets built by user feedback — what have you inferred from these datasets?
When the BYOR bot returns phrases to users, users can rate that suggestion in 3 ways: good, bad or neutral. This is the dataset that our users are thankfully labeling for us, and probably one of the biggest assets that we have. We use this data to improve our suggestion algorithm. For technical readers: we are using this data to train a classifier.
Q What, for you, is the most exciting thing happening in NLP/ML currently?
I’m very excited to see lots of new NLP startups. Medical, legal, shopping helpers, scheduling agents, you name it, there’s an NLP bot. I think this excitement is great. When more people think about these problems, it always increases our potential for innovation.
These days I’m becoming more and more interested in speech recognition technology in NLP. BYOR started as an text prediction app but I think lots of interaction will move to voice from text within a few years.
Q What advice would you give to budding AI developers?
Probably same as the last year. I think success in any arena is always about the fundamentals. To build a good AI system, it greatly helps if you are comfortable with representations and computations in terms of vectors and matrices. Optimization, Regression form the basics will do as well. Also, one should always be very close to the data they are working with.
Another thing is, don’t limit yourself by thinking “I don’t have a degree”, or “I never studied those subjects in school”. If you’d like to build something, start building. Don’t hesitate to invest some time to learn. It doesn’t need to be a formal 2-year program in Machine Learning. There are tons of tutorials written by the brightest minds in the world, all free, available on the web. And thousands of classes (MOOC) from great schools as well. When I was building BYOR, I used a lot of internet tutorials, lectures, Stackexchange, emailing people, talking to people at conferences, etc.
Also, if you don’t understand the math concepts at first, don’t give up. It’s natural that you don’t understand it at first. Don’t get nervous when it takes time to understand. Be patient and keep trying until you get it. From my experience, Math (or whatever subject, but AI is Math) does get easier after a lot of practice, it just takes time.
Lastly, UX matters.
Q Are you excited about speaking at AI With The Best again this year?
Yes, you guys are democratizing expensive AI conferences to the general public. It’s amazing that I can interact with a global audience without traveling. WTB is such an awesome idea, and a great platform.
I’m also excited to attend the talk with other superstar speakers that I admire like Yoshua Bengio, Ian Goodfellow, and Sam Altman! There are also many great practitioners from industry in this conference as well. I think your ticket price is way too reasonable.
Thank you Aerin!!
You can ask Aerin Kim your own questions, and learn more in her talk about using Deep NLP Essentials in 30 mins at our upcoming AI With The Best, Online Developer Conference 29–30th April.
