LLM Deployment & Optimization: Deploying and serving Large Language Models (LLMs) in production environments on GPU clusters (H100, H200, B200). Focus on high-performance inference, deep optimization, and quantization to reduce latency and memory footprint.

Inference Frameworks: Extensive hands-on experience with vLLM, SGLang, and TensorRT-LLM for efficient model serving and throughput optimization.

Infrastructure & Load Balancing: Designing and managing scalable infrastructure for high-load systems. Configuring load balancers (Nginx, Angie, custom SGLang routers) to handle massive request volumes with minimal latency.

High-Performance Computing: Working with cutting-edge NVIDIA GPUs (H100, H200, B200) to maximize hardware utilization for LLM workloads.

Production Stack: Proficient in containerization (Docker), inference servers (Triton Inference Server), and system-level scripting (Bash, Python) for automation and monitoring.

Networking & System Architecture: Deep understanding o

MLOps

PDF Parsing and Vision-Language Model Testing for Catalog Data Extraction

Achievements:

Took ownership of the "PDF parser" direction; researched libraries and approaches, identified promising solutions, and compiled a list of problematic edge cases

Implemented PDF parsing using Camelot on the D'Andrea catalog — the code works, but manual approaches have limitations; recommended advancing the LLM direction

Deployed openbmb/MiniCPM-2.6 in Google Colab; tested table extraction from images

Tested PaddleOCR VL for parsing data from catalogs

Tested Qwen3-VL-4B-Instruct within the client's area of interest

Conducted live testing with the catalog using the Qwen3-VL-235B-Instruct-FP8 model

Tech Stack: Camelot, Google Colab, PaddleOCR, Qwen-VL, MiniCPM, LLMs, Computer Vision

AI / ML Engineer

Completed the "AI/ML Developer" comprehensive course, covering the following key areas:

Python Development: Core syntax, data structures, OOP, and functional programming.

Data Analysis & Visualization: Pandas, NumPy, Matplotlib for data processing and exploratory analysis.

Classical Machine Learning: Regression, classification, clustering, and recommendation systems (scikit-learn).

Deep Learning Frameworks: Building and training neural networks with TensorFlow, Keras, and PyTorch.

Computer Vision: Object detection, image segmentation, OCR, and OpenCV.

Natural Language Processing (NLP): Transformers (BERT, T5), RNN/LSTM, NER, NLTK, and prompt engineering with LangChain.

AutoML: Automated model selection for images, tabular data, and text.

Speech Technologies: Speech recognition and generation.

LLMs & Generative AI: Working with ChatGPT, GigaChat, embeddings, and deploying local state-of-the-art (SoTA) models.

Databases: SQL, PostgreSQL, MS SQL, Oracle.

Big Data & Distributed 

Achievements:

Developed and analyzed a model using FastAI

Implemented a data pipeline via API

Set up an end-to-end data processing workflow within the model

Designed the model architecture

Wrote code to validate predicted recommendations

Tech Stack: Pandas, NumPy, PyTorch, FastAI

Regression Task for Predicting Private House Prices (in conventional units) Based on Input Features

Achievements:

Effectively processed a limited dataset, ensuring high-quality analytical results

Successfully selected and tuned machine learning models for real estate market value prediction

Achieved 93% model accuracy with a mean prediction error of 13,500 conventional units, a significant result considering the price range of hundreds of thousands of conventional units

Tech Stack: Pandas, NumPy, ClassicML, TensorFlow, Matplotlib, scikit-learn

GitHub: https://github.com/Ilyatters/my_project

AI/ML Engineer

LLM deployment on servers. Expertise fine-tuning, quantization, optimization

Know about Ilya Tarasevich's work experience, education, bootcamps, certifications, and much more. Ilya Tarasevich