Machine Learning Data Science

Dmitri Koltsov

email: *********@*****.***, phone: 978-***-****

address: ***** ********* *****, ****** *** Springs, CA 92241 linkedin: https://www.linkedin.com/in/dmitrikoltsov/ github: https://github.com/Dim314159

Summary

Excellent problem-solver with solid theoretical understanding of data science principles, machine learning models and techniques, and AI. Quick learner. Experience

Paragon Research Corporation (Huntsville, AL), AI/ML Engineer, Oct 2024 - current

● Using ML methods for network traffic analysis.

● Technologies and Tools: Data engineering, Clustering, Deep Models, Multi-agent models, Graph NN (PyTorch, Pandas, SciPy etc.)

GiveToGet (Cincinnati, OH), ML/AI Engineer, Nov 2023 - current

● Core Function: Leveraging advanced NLP techniques to enhance the matching accuracy between user profiles and job postings. Focus areas include matching based on location, skills, responsibilities, and previous work experience.

● Technologies and Tools: Proficient use of leading NLP libraries and frameworks, pre-trained Language Models (LMs), AWS Bedrock, AWS Lambdas, etc.

● Model Implementation and Efficiency: Skillfully employing 'last_hidden_state' and

'pooler_output' from pre-trained LLMs to perform nuanced comparisons and cluster the records into categories, which allows for targeted comparisons based on user profile clusters. This method significantly streamlines the matching process, enhancing both the speed and effectiveness of the whole process.

● Data Management: Expertise in parsing and preprocessing text data to create datasets optimized for model input. Efficiently connecting algorithms to AWS DynamoDB, S3 for robust data handling.

● API Development: Developing and maintaining a Flask API to facilitate seamless interactions between the algorithms and the mobile application, ensuring a smooth user experience.

Radical AI (New York, NY), ML/AI squad lead (LLM), Mar 2024 - Oct 2024

● Developing automatic creation of learning materials: using the Google Cloud Gemini model, Langchain, RAG to automatically create quizzes (multiple-choice, fill-in-the-blank etc.) and other learning materials (worksheets, flashcards etc.) from uploaded documents (pdf, docx, csv, url, youtube videos, etc.), enhancing interactive learning tools.

● Utilized Python for backend development, integrating machine learning models with FastAPI to serve web applications.

● Employed prompt engineering techniques to optimize AI model responses, ensuring accuracy and relevance of generated educational content.

● Collaborated in a multi-disciplinary team to implement user interfaces using React, JavaScript, and HTML, improving user experience and engagement. Honda Research Institute (San Jose, CA), Machine Learning Scientist, Jan 2022 - Sep 2023

● Data Collection and Management: Spearheaded the collection of sensor readings from over 100 diverse sensors, meticulously organizing time-series data into structured datasets. This foundational work was crucial for preparing the data for input into machine learning models.

● Data Preprocessing and Feature Engineering: Applied advanced data preprocessing techniques to cleanse the data, extract periods of interest, and conduct comprehensive feature engineering. My efforts in transforming raw sensor data into insightful features were key in enabling accurate model predictions.

● Model Development and Innovation: Pioneered the creation and testing of various machine learning and deep learning models, including CNNs (Convolutional Neural Networks), RNNs (Recurrent Neural Networks), and Transformers, for both classification and regression tasks. My approach to model development was characterized by rigorous experimentation and a commitment to achieving results that were comparable to or surpassed existing industry benchmarks.

● Waymo dataset: Took charge of preprocessing the extensive dataset, ensuring the data was primed for advanced analysis. Crafted algorithms specifically designed to extract complex scenarios involving object behaviors, such as one car following another, overtaking maneuvers, and intersection crossing with yield patterns. These extractions were pivotal in training sophisticated models for future trajectory prediction, significantly contributing to the development of accurate and reliable predictive systems. Skills

● Core Competencies:

Machine Learning: Supervised, Unsupervised, Self-Supervised, Semi-Supervised, XGBoost, KNN, PNN, K-means, Naïve Bayes, SVM, PCA, random forest, bagging, boosting, etc.

Deep learning: Transformers, MLP, RNN, LSTM, CNN, etc.

● Technical Proficiencies:

Python: pandas, NumPy, PyTorch, scikit-learn, LangChain, SpaCy, TensorFlow, Keras, OpenCV matplotlib, SciPy, tsfresh, etc.

AI: LLM, Langchain, RAG, Reinforcement Learning, MCTS, MDP, NLP, EM, GAN, Diffusion Networks, Growing Neural Gas, SOM, etc.

C++, C.

● Supplementary Skills:

AWS, Java, JS, React, FastAPI, Flask, Google Cloud, Azure ML, SAS, SQL, Hadoop, Spark

Education

University of California, Riverside, Master of Science, major: Computer Science Jan 2021 - Mar 2022, GPA: 3.95

University of California, Riverside, Bachelor of Science, major: Computer Science Sep 2018 - Dec 2020, GPA: 3.85

College of the Desert, Associate of Arts, major: Computer Science Jan 2017 - Jun 2018, GPA: 4.00

Relevant Coursework

Machine Learning, Data Science, Deep Learning, AI, NLP, Big Data, Algorithms, Compilers, Statistics

Educational projects

● Chat Bot: Implemented and trained Transformer chatbot with tokenizer.

● Game bot for Backgammon: reinforcement learning with MCTS (value and policy estimation)

● Game bot for Pac-Man game: reinforcement learning (Double Deep Q-Learning)

● Robotic Submarine: Robosub UCR design submarine bot for competition Robosub using ROS ROS

● Currency pair trend prediction: PyTorch, RNN, LSTM, GRU

● Image Analysis: detect age, gender, and ethnicity by face images (PyTorch, CNN)

● Data Challenge in Lawrence Livermore National Laboratory: LLNL, create and train deep learning model to identify potentially hazardous asteroids using celestial imagery

(PyTorch, CNN)

Technical and Leadership Experience

Military: aircraft mechanic, Theater: director, actor, sound and lighting design, Mentor Collective: mentor, Casino: dealer, manager, Yoga: instructor Languages

English (fluent), Russian (native)

Contact this candidate