Matej Kajinic

I’m an engineer, researcher, and venture partner.

My work is driven by a core thesis: the world’s two most valuable currencies are FLOPS (Intelligence) and Watts (Energy).

I have an M.S. in Computer Science (neural networks for financial market forecasting) and a B.S. in Electrical Engineering (electromagnetic waves and real-world propagation models).

Research & Writing
  • ATLAS: Autonomous Task-Learning Agent System (October 2024)
    ATLAS autonomously learns, plans, and executes tasks across business operations while maintaining compliance and security. Source.
  • AI Architectures (October 2024)
    Comparison of AI architectures: Large Language Models (LLMs), Liquid Neural Networks (LNNs), and Neuromorphic AI. Source.
  • GPU Cloud (Internal Paper, October 2024)
    Market overview and updated financial models for GPU cloud infrastructure.
  • Nuclear Energy (July 2024)
    Market overview of nuclear reactor technologies. Source.
  • Energy Is All You Need (June 2024)
    Thesis on why we need more nuclear and solar + battery energy to power AI's growth, guiding infrastructure investments. Source.
  • Economics of GPU (Internal Paper, April 2024)
    Financial models and analysis of GPU cloud economics and market dynamics.
  • The Economics of Intelligence (Internal Paper, March 2024)
    Detailed research on chip supply chain, bottlenecks, and predictions for GPU needs in the near and mid-term future.
  • Application of Neural Networks for Forecasting Financial Market Trends (Master's Thesis, July 2021)
    Built recurrent neural networks (RNN) with long short-term memory (LSTM) for predicting market trends.
  • Educational Simulation Tools for Radio Wave Propagation Models
    International Journal of Education and Information Technologies, Volume 13, 2019.
  • Simulation of Radio Wave Propagation Models on 800 MHz and 1.8 GHz in the City of Dubrovnik
    2018 2nd European Conference on Electrical Engineering and Computer Science (EECS), IEEE, 2018.
Projects & cartography

Technical projects

  • ProntoPR
    A team of specialized autonomous agents that can safely ship incremental features to production using GitHub pull requests, code review, automated tests, and retrieval-augmented generation (RAG). Source.
    Tech: Python
  • AI for Historical Document Analysis
    Natural language processing (NLP) system with long short-term memory (LSTM) to read and analyze handwritten documents from the Dubrovnik archives dating to the 13th century. Source.
    Tech: Python, Tesseract OCR, OpenCV, Pandas
  • Drone Swarm Simulation
    Simulates group drone flights with ROS. Source.
    Tech: C++, ROS, Gazebo, Python
  • Predict Breast Cancer
    Demo test for breast cancer prediction. Source.
    Tech: Python, TensorFlow, Keras

Topographic cartography

Reviewed cartography publications, all available on the Croatian Mountain Rescue Service website:

  • (4) PP Lastovo (2020) - ISBN 978-953-7527-88-4
  • (11) Korcula (2019) - ISBN 978-953-7527-73-0
  • (20) Mljet (2015, 2018) - ISBN 978-953-7527-43-3
  • (25) Dubrovnik (2016) - ISBN 978-953-7527-50-1
  • (28) Konavle (2017) - ISBN 978-953-7527-58-7
  • (30) Dubrovacko Primorje (2018) - ISBN 978-953-7527-60-0
  • (31) Dolina Neretve (2019) - ISBN 978-953-7527-67-9
  • (38) PP Ucka (2019) - ISBN 978-953-7527-91-4
Other