Serhii K. Lead Data Science Engineer /AI & ML Engineer
Summary
- Highly experienced Head of Data Science with 12+ years of experience in creating and managing DS/ML teams in startups and corporate projects;
- Proficient in AI, NLP, Adtech, Fintech, and CV;
- Strong leadership skills and a client-oriented approach;
- Skilled in Python, SQL, Prompt Engineering, HuggingFace, PyTorch, Scikit-learn, Pandas, LangChain, LlamaIndex, Spacy, GPT, Plotly, GCP, AWS, Azure, Postgre, MongoDB, BigQuery, and Vector DB;
- Proactive in implementing innovative approaches for product features using Generative AI, LLM, and GPT;
- Experienced in proposing innovative solutions for new business problems and managing teams;
- Holds a PhD in Data Science and a Master's degree in Computer Science.
Work Experience
Data Science Lead, NDA
Duration: 2021 - now
Responsibilities:
- Implemented a chatbot using GPT LLM and LangChain framework with Rasa framework (to handle user sessions and track chats).
- Chatbot maintains a wellness-style expert conversation with the user and simultaneously analyses what user types for important life events, depressive thoughts, traumas, relational problems, etc.
- Store detected data in the User Profile, and during conversation use the User Profile to answer, taking into account known data about the user.
- It is implemented as a part of ChatBot a RAG system that searches for relevant answers in the wellness database.
- Chatbot handles in a smart way context window size, removes non-important conversation messages from the start or middle, takes into account whether the user is paid or free uses different LLMs, and dynamically creates a prompt.
Technologies: Python, SQL, Prompt Engineering, HuggingFace, PyTorch, Scikit-learn, Pandas, LangChain, LlamaIndex, Spacy, GPT, GCP, AWS, Azure, Postgre, MongoDB, BigQuery, Vector DB, OpenAI GPT, Langchain, Fastapi, Firestore.
Data Science Lead, Model Performance Evaluation
Duration: 2017 - 2021
Summary:
- Managed and participated in team efforts on developing prediction models for CTR, Landing, View-Through, Visibility, and outcome models, such as Visits, Interacted, and Lifted.
- With prediction models achieved >50% cost decrease for running users ad campaigns
Proposed an approach for back-testing and implemented a live-testing framework for evaluating model performance. Researched and developed a CPM optimization model, which decreased bidding costs by 20%.
Responsibilities: Proposed an approach for back-testing and implemented a live-testing framework for evaluating model performance. Researched and developed a CPM optimization model, which decreased bidding costs by 20%.
Technologies: Python, SQL, Prompt Engineering, HuggingFace, PyTorch, Scikit-learn, pandas, LangChain, LlamaIndex, Spacy, GPT, GCP, AWS, Azure, Postgre, MongoDB, BigQuery, Vector DB.
Data Scientist, Business Solutions
Duration: 2016 - 2017
Summary:
- Collaborated with engineering and product development teams;
- Proposed solutions and strategies to business challenges;
- Presented information using data visualization techniques.
Responsibilities: Collaborated with engineering and product development teams. Proposed solutions and strategies to business challenges. Presented information using data visualization techniques.
Technologies: Python, Power BI, Google Cloud.
Machine Learning Engineer & Team Lead, Forex Rates Analysis Engine
Duration: 2015 - 2016
Summary:
- Built engine for technical analysis of forex rates, using modern machine learning and signal processing approaches;
- Presented and constructed cascaded distributed SVM implementation on GPU which increased performance speed by orders of magnitude and removed the memory limitation of classic SVM;
- Implemented high-load multithreaded highly parallelized and low-level optimized client-server applications. Worked closely with remote front-end developers team, and took part in designing interface.
Responsibilities: Built engine for technical analysis of forex rates, using modern machine learning and signal processing approaches. Presented and constructed cascaded distributed SVM implementation on GPU which increased performance speed by orders of magnitude and removed the memory limitation of classic SVM. Implemented high-load multithreaded highly parallelized and low-level optimized client-server applications. Worked closely with remote front-end developers team, and took part in designing interface.
Technologies: Python, STL C++11 concurrency, Boost, ViennaCL.
Machine Learning Engineer, Gesture Recognition Engine
Duration: 2011 - 2015
Summary:
- Successfully commercialized SmartTV and Android engine for gesture recognition;
- Used hardcore classic computer vision approaches for Feature Detection, Optical flow, and Object Tracking for gesture recognition;
- Porting and profiling native applications on Android, using Android NDK and JNI. Developing tools with Qt for collecting and processing big chunks of images and video data, and automated tools for gesture recognition engine testing.
Responsibilities: Successfully commercialized SmartTV and Android engine for gesture recognition. Used hardcore classic computer vision approaches for Feature Detection, Optical flow, and Object Tracking for gesture recognition. Porting and profiling native applications on Android, using Android NDK and JNI. Developing tools with Qt for collecting and processing big chunks of images and video data, and automated tools for gesture recognition engine testing.
Technologies: C++, OpenCV, Android NDK, JNI, Qt.
Education
Data Science PhD
Computer Science Masters Degree
2008 - 2013
Certification
- Power BI
- Google Cloud