Description

I'm Artiom, ML Engineer at KAPTO AI (since November 2024), working on end-to-end machine learning pipelines for document understanding. My core focus is building systems that extract, classify, and reconcile entities across multi-page layouts—combining text classification, NER, and vision-aware approaches.

Technical Foundation

I work primarily in PyTorch. My ensemble and tabular expertise comes from Kaggle-competitive patterns: LightGBM/XGBoost/CatBoost stacking with GBDT calibration layers, OOF validation, and careful fold engineering. For NLP, I'm comfortable with DeBERTa-v3 fine-tuning, DINOv2 embeddings, and the full HuggingFace ecosystem. I optimize defensively—removing fragile invariants before patching, rooting cause analysis in first principles, delivering complete working code over snippets.

Side Project

I'm building a smart cup holder for toothbrushes with an ML model for sound classification deployed on an ESP32 microcontroller—bridging embedded systems and real-time inference constraints.

Industry field of expertise

Languages

English
Fluent
Romanian
Native or bilingual
Russian
Native or bilingual

Workplace preferences

Can work on-site

Washington D.C. (up to 50km)

KAPTO AI
Machine Learning Engineer
LEGAL
January 2024 - Today (2 years and 6 months)
Professional Summary (Jun 2024 – Present)

Motivated Data Scientist & Machine Learning Engineer with nearly 3 years of hands-on experience designing, building, and deploying end-to-end NLP systems for production-grade document processing in enterprise environments (insurance, manufacturing). Strong expertise across the full ML lifecycle, including data preprocessing, model development, evaluation, optimization, and production deployment.

Experienced in system design and large-scale pipeline engineering, with a focus on reducing manual workflows through automation and intelligent document understanding systems. Proven ability to build knowledge-driven architectures, including production-grade knowledge graphs and entity-centric extraction systems. Actively expanding expertise in AI agents, tool-using systems, and MCP-based orchestration frameworks.

Notable technical achievement includes fine-tuning Qwen 2.5 using DeepSEED and QLoRA for entity alignment tasks, successfully deployed into production.

Technical Skills

Programming & ML Frameworks
Python, PyTorch, Hugging Face Transformers, Dagster, Scikit-Learn, Pandas, NumPy, spaCy, LlamaIndex

NLP & Language Models
Text Classification, Named Entity Recognition (NER), Retrieval-Augmented Generation (RAG), LLM Fine-Tuning, Quantization, Distillation, Transformer architectures with custom attention mechanisms

Multimodal & Advanced Models
Qwen 2.5-VL (vision-language), Llama 3, LayoutLM, BERT, GPTQ, AWQ quantization techniques

MLOps & Infrastructure
Scalable ML pipelines, model deployment, monitoring, performance analysis, automated retraining systems

Core Fundamentals
Discrete Mathematics, Probability & Statistics, Linear Algebra, Numerical Methods

Languages
English, Russian, Romanian
Amdaris
Machine Learning Engineer intership
August 2024 - December 2024 (4 months)
Collaborated in a team of interns over two months to develop an MVP platform for investments spanning various economic sectors. Contributed to the Machine Learning pipeline, built robust backend services using FastAPI, containerized the application with Docker, and deployed the final solution on Amazon Web Services (AWS).
FastAPI Pytorch Docker Python React.js