How AI Learned What Matters in an Image
A field atlas tracing how image representation learning evolved from supervised classification to today's self-supervised and latent-prediction methods.
Download PDF →Research and development in Artificial Intelligence.
From underlying architectures and intelligent systems to applications across technical, analytical and creative domains.
About
From foundations to applications
The work combines research into the architecture and behaviour of modern AI systems with the development of agents, reasoning systems and practical applications.
Projects are driven by the problem rather than a fixed domain, combining technical research, computational experimentation and implementation to develop and test ideas across a broad range of applications.
AI Architectures & Model Intelligence
Research into the structure and behaviour of modern AI systems, including architectures, learned representations, model internals and emerging capabilities.
Agentic AI & Reasoning Systems
Research and development of intelligent agents, reasoning systems and multi-agent architectures designed to operate across increasingly complex tasks and environments.
Knowledge & Learning Systems
Developing systems that transform complex information into structured knowledge, supporting research, discovery, learning and deeper human-AI interaction.
Applied Computational Intelligence
Applying AI and computational methods to difficult real-world problems across analytical, technical and creative domains, from quantitative research to entirely new forms of AI-enabled applications.
Research Notes & Papers
Selected work from ongoing research
A selection of research notes, technical papers and longer-form write-ups documenting ideas, experiments and areas of investigation.
Giving Tools to AI Agents
A technical explainer comparing direct scripting, standardized protocols and system prompt configuration as ways of giving AI agents the ability to act.
Download PDF →Open Source AI Models: Weights & Quantization
A guide to how open-weight model files are stored, shared and quantized, and how those choices affect portability, efficiency and performance.
Download PDF →What a Trained Model Actually Looks Like Inside
A close look at the 8.5 billion parameters inside a real trained model, layer by layer, and what their structure reveals about how it was built.
Download PDF →Generative AI in Finance: Market Insight Extraction
A whitepaper on using large language models and retrieval-augmented generation to extract structured insight from large volumes of unstructured market text.
Download PDF →