Sem Leontev

About Sem Leontev

I\'m the creator of SmallML, an open-source Bayesian machine learning framework that enables accurate predictions with small datasets (50-200 observations). This solves a critical barrier that prevents 90% of U.S. businesses from leveraging Machine Learning.Traditional machine learning requires data points for reliable predictions. Small and medium-sized enterprises (SMEs) typically have only 50-500 customers, causing standard ML algorithms to fail. This \'small-data problem\' locks 33 million U.S. SMEs out of the AI revolution, despite having critical business needs like churn prediction, fraud detection, and demand forecasting.SmallML achieves enterprise-level accuracy with minimal data through a three-layer architecture combining transfer learning, hierarchical Bayesian modeling, and conformal prediction. By transferring knowledge from large enterprise datasets and pooling statistical strength across multiple businesses, the framework compensates for individual data scarcity and provides robust uncertainty quantification.My work on SmallML is independent research is democratizing machine learning for resource-constrained businesses. If you\'re interested in small-data machine learning, Bayesian methods, or SME analytics, I\'d love to connect and discuss potential applications or collaborations.

Experience

1. Data Scientist

   Bio-Rad Laboratories

   Jan 2024 — Present · Irvine, CA, US

   Analyzed unstructured work order text data using an NLP model (SpaCy) to predict root causes of operational issues. Provided actionable insights that informed process improvements and enhanced operational efficiency.• Developed and deployed an AI chatbot with Retrieval-Augmented Generation (RAG) using LLM, Snowflake and Streamlit, enhancing user interactions and response accuracy.• Transformed static Excel financial models into a dynamic forecasting system within Snowflake. Developed a time-series model to project future financial outcomes and designed a Power BI dashboard with customizable parameters, allowing Director-level users to simulate various economic conditions and strategic interventions.

Education

• UC Irvine

  Master of Science - MS, Business Analytics

• Peter the Great St.Petersburg Polytechnic University

  Electrical and Electronics Engineering