Derek Hommel

Machine Learning/Natural Language Processing Researcher

master of computational linguistics, SNU 2018

contact

currently living in Seoul, South Korea.

please contact me via email at derekshommel at gmail.com or through LinkedIn

message me from Korea at 010 - 9482 - four two zero eight

technical skills

• languages : python
• deep learning libraries : tensorflow, keras, pytorch
• engineering : amazon aws, docker, kafka, tensorflow-serving

professional projects

listed in chronological order of participation, earliest to latest

retrieval-based dialog assistant and conversation classifier

clients: korean e-commerce startup, international bank

• data analysis of 1M+ historical chats
• ML models for classifying at sentence and dialog level (dialog level: 83 F1 score)
• conversation response recommendation
• worked with technical advisors, engineering team to guide project development
• mentored company interns and advised them on project sub-tasks
• converted models & system from Korean to English for second client in new domain

named entity recognition/joint SLU system

client: korean e-commerce company (with Seoul Innovation Challenge Grant)

• researched and implemented a flexible neural NER/joint-SLU tool for English and Korean
• led a team of two data analyst interns and two junior ML software engineers to complete project
• korean grocery shopping domain entity (product, brand, size/type, count) detection (recall: 99.4)
• english travel domain using ATIS corpus (intent accuracy: 96.3, slot F1: 93.71)

limited-domain neural machine translation

client: joint venture with VC firm

• supervised one junior ML software engineer
• worked on data bootstrapping (webcrawling) and processing; data input methods
• researched seq2seq translation methods, transfer learning for low-resource translations
• trained custom translation model (BLEU: 28.64, compared to Google Translate BLEU: 2.05)

NLU model serving for large-scale deployment

client: 예스24 (Yes24)

• created CLI-based model trainer and exporter to train and save tensorflow models for deployment
• created and deployed lightweight REST-based model server
• configured and deployed model server with Docker
• worked with ASR team to integrate NLU with STT for voice based search
• currently available on Yes24 mobile app (as of Nov 2019)

Korean neural text-to-speech

client: internal service

• tested various TTS and vocoder models for best performance
• created multi-speaker TTS model with Tacotron 2 and WaveGlow for Korean language
• currently researching alternate vocoders and frameworks to improve inference speed

personal projects

MLPy
[ongoing] education-focused implementation of machine learning algorithms in python+numpy with detailed comments

Magic The Gathering Card Generation with LSTM Language Modeling
[ongoing] a fun side-project for generating MtG cards using neural networks.
see some cards at my MtG Cardsmith page

Named Entity Recognition with Bidirectional LSTM-CRF in Keras
a keras implementation of word-level bi-LSTM-CRF network for entity extraction

Multipurpose LSTM_CRF Network with Attention for NER & Intent Detection
an extension of the above project for a unified NER and intent detection network

work experience

Principal NLP Researcher (March 2017 – Present)
Atlas Labs, Seoul, Korea
assisting in researching and developing AI & NLP technologies for dialog assistant software.

Guest English Teacher, English Program in Korea (EPIK) (2012 – 2014)
Ohyun High School, Jeju Girl’s High School & Jeju Girl’s Commercial High School, Jeju City, South Korea
developed syllabus and curriculum, and taught weekly conversational English classes to high school students.
volunteered to lead student-driven extracurricular activities including persuasive writing & magazine publishing club, debate club, and programming/game development club.

Fulbright ETA Orientation Coordinator (2010)
Jungwon University, Goesan-gun, South Korea
conducted intensive seven-week orientation and training program covering Korean educational system, teaching methodology, and cultural adjustment. primary liason for maintaining site logistics with university.

Fulbright English Teaching Assistant (ETA) (2009 – 2012)
Aphae High School, Jeollanamdo, South Korea
Daejeon Yongsan High School, Daejeon, South Korea
developed syllabus and curriculum, and taught weekly conversational English classes to high school students.

education

Seoul National University (2015 – 2018)
Master’s in Computational Linguistics
thesis: Data-Driven Input Feature Augmentation for Named Entity Recognition

University of Rochester (2006 – 2009)
Bachelor of Arts magna cum laude in Japanese Language in Culture
with minor in Linguistics

Rochester Institute of Technology (2004 – 2005)
Software Engineering on the RIT Computing Medal scholarship

awards & honors

Naver AI Hackathon 2018
Round 3/Finalist with Team Atlas

NIIED Korean Government Scholarship Program Graduate Scholarship (2014)
Seoul National University – Linguistics

Test of Proficiency in Korean (TOPIK) (2014)
Level 4 Proficiency

Teaching English as a Foreign Language (TEFL) (2012)
100-hour Certificate

Fulbright Scholar (2009)
English Teaching Assistant grant to South Korea