I'm an ML/AI Engineer with 3.5+ years of experience building production-oriented machine learning systems across computer vision, time-series modeling, and LLM applications.

This profile contains selected projects, courses I designed and authored, and open-source contributions related to:

• agentic LLM applications, RAG, and evaluation;
• FastAPI-based ML and LLM services;
• from-scratch computer vision architectures, image segmentation, and seismic denoising;
• Python concurrency and backend engineering.

Most of the production systems I have worked on are proprietary and cannot be published due to company confidentiality policies. My production contributions are maintained in internal GitLab repositories. The projects presented here demonstrate similar engineering approaches through independent applications, educational materials, and open-source contributions.

An AI-powered math tutor designed to guide users through problem solving rather than simply return an answer.

The project includes:

• LangGraph-based workflows for hints, guided solutions, and answer evaluation;
• a FastAPI backend;
• a PostgreSQL-backed repository of solved problems;
• support for open-source model serving with vLLM, TGI, and Ollama.

Tech stack: Python, FastAPI, LangChain/LangGraph, PostgreSQL, vLLM, TGI, Ollama

A LangGraph-based assistant for university applicants and students.

The project combines program matching, required-document checklists, and a RAG assistant over official university admission rules and educational documents.

Tech stack: Python, FastAPI, LangChain/LangGraph, RAG, PostgreSQL

A practical LLM engineering course that I designed and taught at ITMO University. It covers:

• transformer architecture, KV cache, quantization, and inference optimization;
• open-source model serving with vLLM, TGI, and Ollama;
• LangChain and LangGraph workflows;
• FastAPI-based LLM services;
• interactive user interfaces for LLM applications with Gradio;
• tracing, observability, and evaluation with Langfuse;
• agentic applications and production-oriented LLM system design.

The repository contains course presentations, practical materials, runnable examples, and exercises covering the development, evaluation, and deployment of complete LLM applications.

A practical Python course that I designed and delivered internally at Gazprom Neft. It includes presentations, notebooks, runnable examples, quizzes, and exercises covering:

• operating-system and CPU fundamentals;
• multithreading and synchronization;
• multiprocessing, IPC, and process pools;
• asynchronous programming with asyncio;
• networking, WSGI, ASGI, and FastAPI-based asynchronous services.

From-scratch implementations of core convolutional neural network architectures, accompanied by experiments exploring their internal design and behavior. One of the implemented architectures was applied to build and evaluate a seismic data denoising solution.

I have contributed to BatchFlow, an open-source Python framework for machine learning pipelines, including computer vision tutorials, framework fixes, and tools for extracting intermediate neural network activations.

• LinkedIn
• Email: estrievich99@gmail.com