Research

Research and experimentation across AI and emerging technology, from local language models to game development and hardware.

  1. Since 2025

    Offline LLMs

    Building a permanent offline environment for running and storing language models without internet access, APIs, subscriptions, or cloud services.

    We use open weight models with llama.cpp, keep them on dedicated local storage, and maintain offline backups so the environment can be moved between machines.

    We test models on limited hardware and document their practical capabilities and limitations.

  2. Since 2025

    Small Language Models

    Building small language models from scratch to better understand how they work.

    The work covers datasets, tokenization, model architecture, training, inference, quantization, and evaluation.

    When local hardware is insufficient, we use rented GPU infrastructure for training and benchmarking while keeping the development environment, datasets, code, and results under our control.

  3. Since 2026

    Game Worlds

    Exploring how language models can be used inside 2D game worlds.

    The work focuses on character behaviour and decision making, allowing characters to respond to their surroundings and each other rather than relying entirely on fixed scripts.

  4. Since 2026

    Robotics & Hardware

    Experimenting with language models alongside small hardware projects.

    The work explores how local models can interact with sensors and physical hardware, including simple autonomous behaviours.

  5. Since 2026

    Learning & Reasoning

    Researching how models learn, reason, and arrive at their outputs.

    We use small models and interpretability techniques to investigate internal processes rather than only evaluating what a model produces.

    We’ve also developed ways to measure and test parts of this process, and continue refining them through experiments.

  6. Since 2025

    Autonomous Rescue Swarms

    Researching and developing an autonomous AI swarm designed to assist people during emergencies and crises.

    The work covers multiagent systems, robotics, cybersecurity, autonomous decision making, and coordinated response.

    The focus is on safe, reliable technology that can identify hazards, assess situations, and carry out rescue operations with minimal human intervention.