Looking for a job

George Maltsev

@georgemaltsev

Senior Machine Learning Engineer at UBS, improving enterprise RAG search and building a Deep Research Agent.

At UBS, I rebuilt an on-premises RAG platform for internal documents, improving Recall@5 from 66% to 90%. I also redesigned retrieval to cut average knowledge-search time from about 2.5 minutes to 25 seconds.

I designed and developed a production Deep Research Agent with LangGraph, bringing together query planning, iterative retrieval, and evidence-based response generation. It reduced median research time-to-answer by about 50%.

At Siemens, I led three engineers developing a RAG assistant for railway engineering documents, increasing Recall@5 from 70% to 89%. I also independently developed an agentic RAG system for bid and solution engineers that reduced average task completion time from about 24 minutes to 11 minutes.

At Yandex, I improved developer documentation search and flight refund-risk prediction, including raising search Recall@5 from 64% to 87% and model ROC-AUC from 0.69 to 0.82. Earlier, as a Data Science Intern at Meituan and ByteDance, I worked on delivery-time prediction and real-time data pipelines for recommendation systems.

Experience

Work history, roles, and key accomplishments

Current

Senior Machine Learning Engineer

Current

UBS

May 2025 - Present (1 year 5 months)

Enterprise RAG Platform: Rebuilt an on-premises Retrieval-Augmented Generation platform covering approximately 180K internal documents. Designed hybrid retrieval pipelines combining BM25 and dense embeddings, integrated Qwen3 reranking, and implemented document-level access controls and revision-aware filtering. Improved Recall@5 from 66% to 90%, significantly increasing the accuracy and relevance

Senior Machine Learning Engineer

Siemens

Oct 2023 - Mar 2025 (1 year 5 months)

Railway Engineering RAG Assistant: Led a team of three engineers in developing an enterprise RAG assistant for approximately 90K railway engineering and technical documents. Designed hybrid search combining BM25, semantic embeddings, and reranking to help engineers efficiently retrieve technical specifications and engineering knowledge. Increased Recall@5 from 70% to 89%.

Agentic RAG for Bid Engi

Middle Data Scientist

Yandex

Jun 2022 - Sep 2023 (1 year 3 months)

Developer Documentation Search: Led the redesign of documentation search for developers integrating Yandex APIs and services, covering approximately 65K technical documents.

Rebuilt the retrieval pipeline from dense-only search to hybrid BM25 + dense retrieval, introducing multilingual reranking and section-aware chunking.

Search Quality and Optimization: Improved Recall@5 from 64% to 87%, incr

Junior Data Scientist

Yandex

May 2021 - Jun 2022 (1 year 1 month)

Flight Refund Risk Prediction: Developed and deployed machine learning models for Yandex Travel using approximately 350K historical flight bookings.

Rebuilt the existing refund-risk prediction pipeline, replacing Logistic Regression with CatBoost and improving feature engineering and model validation.

Predictive Modeling and Feature Engineering: Analyzed booking behavior and refund patterns usi

Data Science Intern

Meituan

Jan 2021 - Apr 2021 (3 months)

Developed and evaluated machine learning models for predicting food delivery times using historical orders, courier activity, restaurant characteristics, and location-based features.

Built data preprocessing and feature engineering pipelines using Python, SQL, Pandas, and Spark, preparing structured datasets for model training and experimentation.

Trained and compared regression models using XGB

Data Science Intern

ByteDance

Sep 2020 - Dec 2020 (3 months)

Contributed to the development of real-time data pipelines supporting recommendation systems, focusing on processing user interaction events and generating behavioral features.

Implemented and optimized data processing tasks using Python, SQL, Apache Kafka, and Apache Flink to aggregate user activity for downstream ML workflows.

Developed data validation checks and assisted in debugging streamin