Meet Aerin Kim, Data Scientist turned Co-Founder of startup BYOR (Build Your Own Resume), building an AI-based resume helper using NLP parsing. When a user uploads their resumé on the web-app, it gives suggestions on how to improve your resume regarding its wording or phrases. Catered for Data scientist, analysts and engineers, we find out how this works.
WTB. Please tell us a bit about your background prior to BYOR and how you get into data?
I was a NLP data scientist at a startup called Boxfish. I did a lot of Twitter text modelling there and had been fascinated every day by the amount of information that could be gleaned from all the text that people were generating. Because it was a startup, our team was building the product from scratch over many iterations. That training helped me later when I turned my idea into a product (BYOR).
WTB. What propelled you to push NLP parsing technology for Resumés?
My co-founder and I have been volunteering as resume reviewers and mentors for Columbia University since 2014. Every year, we found there is a pattern for weak resumes and we found ourselves giving students the same advice year after year. We saw an opportunity for some automation in this resume reviewing process.
Also at college career centers, it’s hard to get a one-on-one session with career advisors because the student-to-advisor ratio is hundreds to one. We decided to create a tool that could be used by students to review their resume prior to meeting their career advisors, or as a substitute.
The BYOR project started as the class project for the CS 224d (Dr. Richard Socher) at Stanford. Rohit and I took that class online.
WTB. How do you train the word embedding neural networks to find similarities and relations between phrases?
The main way to find similarities and relations between two different phrases is converting them to phrase vectors and then finding the distance between these vectors. There are many different ways to calculate phrase vectors. The simplest way that anyone can try is to first train the word vectors and then weight average those word vectors used in the phrases.
WTB. What can BYOR do compared to other CV checkers?
Currently, there is no company that suggests result phrases on a specific sentence. Even AI companies with good amount of funding don’t open their platforms like us. Inviting people to upload any kind of resume and give them suggestions is a challenging problem on many levels and taking it on requires a little bravery.
What traditional CV checkers do is simple keyword extraction or keyword counting to check whether certain words are used or not. They don’t understand the user’s resume line by line semantically.
WTB. What’s been the most exciting part of your startup adventure?
The most exciting part is when we improve the “phrase suggestion algorithm” day by day and succeed in generating phrases that make sense.
Also, before the startup, I used to work for a big bank. If you are an employee of a big company, your job description is very narrowly focused. But in a startup, I can experiment with all parts of the product. It has been very exciting for me so far.
Also, it’s amazing to see many people contributing to BYOR voluntarily.
WTB. If it’s not a secret, which is your favourite technological setup?
It’s not a secret. We use python django for web. All NLP/deep learning code is written in python.
To train word vectors, we use code written in C.
WTB. What advice would you give to budding AI developers?
If you are AI developer, Applied Math basics are very important for you. Invest some of your time to go over Linear Algebra, Optimization, Probability that you learned during college.
WTB. Are you excited about speaking at September’s AI WTB?
Yes! I like that you guys priced it under 100 bucks so that general public can attend. And it’s on the internet!!! I think you guys are doing the right thing. People/students shouldn’t have to have sponsors to attend such tech conferences. With The Best line-up is as good as a $3000 conference.
Thanks so much Aerin — looking forward to your talk!
Aerin will be speaking about “Phrase2vec in practice” on Sunday 25th 6pm and Co-Founder Rohit Pandey (ex-LinkedIn) on the Microsoft Azure team will also be speaking at 4.40pm about “Using entropy as a metric for anomaly detection in categorical histogram” at ai.withthebest.com
Originally posted to Medium/Withthebest