Andrii Ushakov

A four-stage generative AI pipeline in Python and PyTorch: Qwen3-TTS emotion-prompted speech synthesis, Vevo voice conversion with vocal separation, Qwen-Image-Edit to adjust the reference face's expression, and MEMO memory-guided diffusion rendering the talking-head video with a target emotion. Manifest-driven CLI with a reproducible pixi environment; MEMO and Amphion forked and refactored into importable, installable modules.

A full-featured interactive quiz application engineered with Angular 15, Angular Material, strict TypeScript, RxJS and Firebase (Firestore, Cloud Functions, Hosting). Features a dynamic quiz builder and player, 100% line coverage with Jest enforced by GitHub Actions and Codecov, and shared code split into four published npm packages.

Ten merged pull requests to a PyTorch codebase for cross-domain mosquito species classification from wingbeat audio: streaming dataset pipeline (MosaicML Streaming), Pydantic-validated experiment configuration with automatic trial-grid generation, feature caching by content hash, switchable loss functions, ROCm training environment and an Optuna hyperparameter-optimization PR.

An Angular 19 standalone application with Angular Material, signals interop and built-in control flow: trip constructor with flights, stays and attractions, trip search and publishing, public profiles, comments and notifications. Integrates a REST API through HttpClient with a functional Bearer-token interceptor and route guards.

A TypeScript / Node.js CLI framework that streamlines GCode post-processing for PrusaSlicer: G-code parser and serializer, composable pipe modules, toolhead and part-cooling-fan emulation, and a built-in fan-optimizer module that smooths rapid fan-speed changes. Webpack build with a cross-platform installer.

A full-stack quiz platform built with React 17, Firebase 9, and Material UI v5. Features Redux-managed state, and a Firebase Cloud Functions backend. Secured with Firebase App Check and reCAPTCHA v3. Ships with a service worker for offline-capable delivery.

A C++ Arduino library for on-the-fly web page generation and self-hosting on MCUs. Utilizes a highly modular component architecture with Materialize CSS styling directly served from low-memory embedded devices.

A C++ Arduino library that automates the ESP32 camera setup routine and exposes an easy way to tune camera settings for the most common boards (AI Thinker, Wrover Kit, M5Stack). Published on the PlatformIO registry.