Senior Research Engineer (Machine Learning, NLP)
2019 - 2020Abto Software- Implemented a chatbot model designed for stable progressive learning for five years.
- Reimplemented part of Rasa chatbot framework to fix OOM error and make it production-ready.
- Found and fixed OOM error and experiment design problem in SoTA intent recognition paper.
Technologies: Docker, Rasa.ai, Scikit-learn, TensorFlow, PythonSenior Research Engineer (Machine Learning, NLP)
2017 - 2019Deloitte- Implemented a contract classification algorithm with 95% accuracy which outperformed humans.
- Developed a complex text summarization algorithm with 80% accuracy.
- Created a company classification algorithm which increased human analytics quality by two times.
- Completed all stages of complex text summarization problem including framing the problem, annotating the data, creating the model, deploying and supporting.
- Organized data science hackathon which resulted in three new hires.
Technologies: NLTK, Keras, TensorFlow, Scikit-learn, Machine Learning, Deep Learning, Data Science, Java, PythonResearch Engineer (Machine Learning, NLP)
2013 - 2017Grammarly- Developed a new sentence splitting component with 2X improvement in error rate.
- Built an auto-correct feature with 99% accuracy compared to 90% existing version.
- Implemented high-load storage with approximately 1,000,000 key-value requests per second per server.
- Reimplemented contextual spell checker with 10x speed-up.
- Implemented a practical solution with 99,9% specificity and 30% recall to an open problem of detecting sentences which don't make sense.
- Processed petabyte-scale internet dump using Apache Spark for creating n-gram language models.
- Prepared and conducted two courses on algorithms and data science.
- Increased part of speech tagger quality from 96% to 97%.
Technologies: Amazon Web Services (AWS), OpenNLP, Keras, TensorFlow, Scikit-learn, Deep Learning, Natural Language Processing (NLP), Data Science, Java, PythonAnalyst Engineer (Machine Learning)
2011 - 2013UkrNet- Developed an anti-spam system for an email provider that had tens of millions of users.
- Created a geo-tracking solution for ads which increased accuracy from 50% to 80%.
- Implemented Nginx plugin using C with 6000 geo-tracking requests per second per server.
- Built the back end for email IMAP protocol which enabled 100,000 users to use their favorite email clients.
Technologies: Natural Language Processing (NLP), Machine Learning, Weka, MySQL, C++, Java