Sezim Yertanatov

@yersezim

ML Engineer focused on LLMs, NLP, and Applied AI

I'm an ML Engineer with a background in Computer Science from Harvard University, where I earned both Bachelor's and Master's degrees through an accelerated program.

Open to

Finding a mentorJoining pet projectsProject collaborationPaper discussionsHiring & referrals

About

My recent work has focused on building production LLM systems, including RAG pipelines, local model deployment with vLLM, and large-scale ETL pipelines for sensitive data. I'm particularly interested in NLP, efficient LLM inference, and designing scalable AI infrastructure that brings machine learning into real-world products.

Career

Experience: 1 years

Experience

ML Engineer

Financial Monitoring Agency of the Republic of Kazakhstan

Jan 2026 — Jul 2026

• Developed a hybrid retrieval-augmented generation (RAG) pipeline with reranking for an internal knowledge base. • Deployed local language models with vLLM and optimized inference performance, resource utilization, and throughput. • Applied graph algorithms, including Yen’s k-shortest paths algorithm, to optimize analysis of internal datasets. • Built end-to-end ETL and ELT pipelines to process terabyte-scale sensitive data using PostgreSQL and ClickHouse, leveraging prompt-engineered LLMs for data enrichment and analysis.

Education

BA, MS in Comp Science

Harvard University

2021 — 2025

Relevant Coursework: Intro to Algorithms and their Limitations; Data Structures and Algorithms; Systems Programming and Machine Organization; High Performance Computing; Machine Learning; Introduction to Probability; Linear Algebra and Differential Equations; Distributed Systems; Systems Security.

Selected projects

Multi-Objective Explanation Tuning (METRIX)

• Created a novel method that jointly optimizes infidelity and sensitivity in model interpretability (ImageNet, MNIST). • Achieved 99.9% lower sensitivity than baseline without sacrificing model faithfulness; visualized 200+ class explanations. • First to propose a closed-form update rule for infidelity-sensitivity balancing in feature attribution.

Open project

Skills

PythonMLData EngineeringFastAPIPyTorchC++Distributed Systems