👋 Hi, I’m Tomas

I build intelligent systems that bring structure to unstructured data.

As a Machine Learning Engineer and AI Consultant, I work across computer vision, natural language processing, and audio, often combining them into multimodal solutions. Whether optimizing document workflows that handle 30k+ daily financial records or designing AI that connects video to music, my goal is to craft robust, scalable systems that work in production.

🔬 What I’m Working On

• Document Intelligence: Improving classification and information extraction in multilingual financial pipelines using Transformers, semantic search, and active learning.
• Face Analysis Toolkit: Creator of [facetorch](https://github.com/tomas-gajarsky/facetorch), a modular PyTorch library for facial detection, representation, and deepfake detection, downloaded over 40k times.
• Multimodal ML: From aligning music embeddings with video content to designing session-based recommenders using RL, I’m passionate about how perception and semantics intersect.

💡 Why Sponsor Me?

Your support allows me to:

• Continue building and maintaining open-source tools like facetorch.
• Explore under-researched challenges such as imbalanced learning and long-tailed distributions in neural network training.
• Develop new libraries that abstract complex ML workflows into reproducible, user-friendly packages.

I’m especially focused on bridging the gap between research and production. Sponsorship helps me invest time into clean design, documentation, and long-term maintainability, not just experiments.

🧠 Tech Stack

Languages & ML Tools: Python, PyTorch, Hugging Face, Scikit-learn, SQL
Infra & MLOps: Triton Inference Server, Azure ML, MLflow, Docker, ONNX
Data Systems: Elasticsearch, Qdrant, LangChain
Focus Areas: Transformers, LLMs, CV, NLP, Audio, AI Agents, Semantic Search

🤝 Let’s Connect

I’m open to collaboration, questions about my work, or shared exploration of new AI frontiers. Sponsors get early access to ideas and in-progress tools — and play a direct role in shaping what's next.

Thanks for reading — and supporting independent AI development.