NLP/ML Developer

Описание

Nikita Stepanov
 t.me/nikitast nikitast.diff@gmail.com 
n1kstep +7(910)-099-8570
EDUCATION
Bauman Moscow State Technical University Moscow, Russia
Bachelor’s in Mathematics and Computer Science September 2020 – August 2024
GPA 4.6/5.0
EXPERIENCE
Tinkoff AI Center Moscow, Russia
R&D NLP Intern February 2022 - May 2022
○␣
Collected first labelled Russian Dataset of Toxicity in Context; created crowd-sourcing markup: experimented on questions’
form and tasks for raters’ training and control to improve agreement metrics such as Kohen’s kappa and Krippendorff’s
alpha
○␣
Experimented on fine-tuning BERT for Toxicity in Context classification using HF Transformers; optimized (JIT, Torch-
Script) best model and prepared for deployment using TorchServe
○␣
Tools : Python, PyTorch, TorchServe, Transformers, W&B, Git, Docker, Kubernetes
Sberbank Moscow, Russia
Data Science Intern October 2021 - February 2022
○␣
Assisted in feature extraction from encoded data, audio matching using OpenMP; parsed data from open web resources
for additional model features
○␣
Experimented with Dimensionality Reduction methods (PCA, UMAP) and Classical ML algorithms; implemented Sber
LightAutoML for the research; achieved 0.91 ROC-AUC on time series with table metadata
○␣
Tools : Python, PyTorch, Sklearn, LightAutoML, Vosk, C++
PROJECTS
Language Classifier | Python | n1kstep/lang-classifier
May 2022
○␣
Trained XLM-Roberta model to detect the language of the text and to segment text into parts that belong to different
languages. Created algorithm for multilingual train data generation. Deployed models on
HuggingFace
Style Transfer Bot | Python | n1kstep/style-transfer-tg-bot
May 2021 - June 2021
○␣
Implemented Style Transfer model based on ”A Neural Algorithm of Artistic Style” (Gatys, Ecker, Bethge)
COMPETITIONS
Evraz AI, Data Science track (team competition) October 2021
○␣
Finalists: 5th out of 113 teams, ML Botalka Team
Codeforces Raif ML Round, Football Betting June 2021
○␣
Prizewinner: top 10% (54th out of 517 participants)
EDUCATIONAL PROGRAMS
Tinkoff Generation ML/DL Courses February 2021 – December 2021
○␣
2nd out of 80 students in the perfomance table (ML), assisted in creation of ML Course cheat sheets
Deep Learning School Course October 2020 - January 2021
○␣
Participated in InClass Competition: Journey to Springfield, top 8% in the Leaderboard
TECHNICAL SKILLS
○␣
Machine Learning, Deep Learning, Software Development
○␣
Programming Languages and Tools: Python, C++, Bash, Git, Docker, Kubernetes
○␣
Libraries and Frameworks: PyTorch, TorchServe, HuggingFace, Sklearn, NumPy, Pandas, OpenCV