— |
I design, train, and deploy production-grade ML systems — from transformer models to real-time inference at scale.
18+ years building end-to-end software systems
I'm an ML Engineer and open-source developer from Podolsk, Russia, focused on practical AI: local, offline-first systems, deep learning research, and production software. I specialize in NLP, RAG pipelines, and MLOps — turning ideas into reliable systems that run without external dependencies.
I believe the best AI is the one you can run yourself. My work spans an air-gapped AI workstation for case investigation (CaseAI), philosophically-inspired loss functions for PyTorch, and a Russian-market restaurant management system — always with reproducible, testable engineering behind it.
Tools I use to ship models that matter
A timeline of shipping ML to production
Leading software for Chinese automotive platforms, including NDA projects for Chinese automotive brands. AI voice interfaces, backend services, automotive Android/Linux integrations and diagnostics. Managing a distributed team of up to 20 people.
Researching, diagnosing and modifying Linux-based automotive computer systems, navigation and vehicle software stacks. Published technical reverse-engineering and repair material for the engineering community.
Backend, frontend, REST APIs, relational databases, automation, integrations, Linux deployment and production debugging. Repeatedly moved across unfamiliar technology stacks to solve end-to-end product problems.
Selected open-source work — from research to production
Fully local AI workstation for case investigation: OCR, Whisper audio transcription, hybrid RAG search, and Fast 32B / Deep 70B LLMs. Air-gapped on Ubuntu 24.04 + NVIDIA — zero external APIs.
Experimental PyTorch add-on: philosophically motivated yet technically verifiable loss functions and regularizers — epoche, hermeneutic circle, dialectics, perspectivism, virtues, semiosis. 45 passing tests.
Open-source restaurant management system for the Russian market — an iiko alternative built on ERPNext/Frappe with Russian integrations: 1C, 54-FZ fiscalization, acquiring, delivery aggregators, EGAIS and Mercury.
Live from github.com/webzuweb
URY - FOSS restaurant management system super powered by ERPNext.
CodeKey - ESP32-S3 USB HID keyboard bridge (Android controller)
Философски мотивированные, но технически проверяемые функции потерь и регуляризаторы для …
AI рабочая станция следователя с локальным ИИ DeepSeek: OCR, аудио, гибридный RAG-поиск, …
A community-maintained Python framework for creating mathematical animations.
Backup tool for ClickHouse DBMS
Rust implementation of behavior trees for deterministic AI (now with Python bindings)
Copper is an operating system for robots - build, run, and replay your entire robot dete…
Python package for the evaluation of odometry and SLAM
Android phone -> ROS2 sensor bridge for Jetson Nano (IMU/GPS/camera/telemetry) · Мост «An…
A fast, scalable, high performance Gradient Boosting on Decision Trees library, used for …
GigaAgent — это универсальный агент-оркестратор для решения широкого круга задач (ReAct +…
Official repository for IPython itself. Other repos in the IPython organization contain t…
🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in…
Deep universal probabilistic programming with Python and PyTorch
Open standard for machine learning interoperability
A unified framework for machine learning with time series
NLTK Source
A python library for user-friendly forecasting and anomaly detection on time series.
I'm always open to discussing ML research, open-source collaboration, or new opportunities.