I build intelligent systems — from fine-tuning LLMs and training audio models to deploying production ML pipelines. 7+ years of software engineering with a focus on bringing research to real-world applications.
I started my career in .NET enterprise development — building CRM integrations, reporting systems, and Azure-hosted applications for clients in banking, automotive, and tourism.
Over time, I shifted toward machine learning. Today I work on LLM fine-tuning, transformer-based audio models, and ML classification tools.
I write about my experiments on Medium and I'm always looking for interesting problems at the intersection of software engineering and machine learning.
Training and adapting large language models for specific domains. Local deployment and optimization for production use.
Audio ML pipelines — classification, generation. Transformer-based architectures for audio.
Production-grade APIs and cloud infrastructure. Strong .NET ecosystem background with Azure services.
Strong foundation in software development — clean code, algorithms, data structures, and object-oriented design.
Fine-tuning local LLMs for educational apps. ML classification tools for audio engineers. Conversational chatbots for Czech e-commerce.
Built a reporting SPA from scratch. Domain-driven design backend hosted on Azure.
Tourism prediction app backend. NetLogo simulation integration with university researchers.
Banking CRM integrations. Azure Functions, SSRS reports, MS Dynamics customization.
MS Dynamics CRM plugin development. Customer requirements analysis.
Attendance system in PHP/Nette. WinForms maintenance app in C#.
ML classification tools for audio engineers using transformer-based models. Architecture experiments to improve production accuracy.
Fine-tuning local LLMs for an educational app for older adults. Recommendation algorithms and React frontend.
4+ years building strategies with Freqtrade. GBDT models via FreqAI for crypto market prediction and automated decisions.
Conversational chatbot for a Czech e-shop recommending products using a local LLM with optimized prompt engineering.
Use case: Visualized audio embeddings help us build balanced datasets by revealing class distributions and identifying underrepresented samples.
Each point represents a speech sample from one of 7 European languages. Audio files are converted into 768-dimensional embeddings using a ContentVec transformer model, then projected into 3D space with UMAP. The resulting clusters show that the model captures language-specific acoustic patterns — similar languages appear closer together.
Open to ML engineering roles, freelance projects, and interesting collaborations.