Opaque decisions
Reasoning paths are rarely accessible in a meaningful form.
Independent research laboratory
Building the computational foundations of explainable intelligence.
We research alternative approaches to artificial intelligence where transparency, transferable representations, and principled computation are built in from the start.
CORRESPONDENCE / 01
MATCH / TRACE / VERIFY
The challenge
Capability without a clear account of why.
Many high-performing systems arrive at an answer through representations that remain difficult to inspect. This creates practical limits for verification, scientific use, and work across domains.
Hidden representations
Similarity is encoded without an account of what corresponds.
Fragile transfer
Knowledge does not always travel cleanly between domains.
Limited verification
Claims can be difficult to trace, test, and contest.
Research philosophy
Intelligence should not only be powerful—it should also be understandable.
We investigate models where reasoning can be inspected, traced, and explained through explicit structure and correspondence, rather than inferred solely from hidden states.
Research areas
Questions worth making legible.
Explainable Artificial Intelligence
Systems whose decisions can be examined in context.
Representation Learning
Structured, transferable forms of knowledge.
Structural Correspondence Search
Finding meaningful matches across representations.
Cross-Domain Intelligence
Reasoning that carries its structure between fields.
Computational Models of Reasoning
Explicit accounts of search, inference, and comparison.
Robotics
Embodied systems with inspectable behavior.
Bioinformatics
Computational tools for interpretable biological inquiry.
Scientific Discovery
Methods that make patterns available to investigation.
Featured research
HUNTER
Hungarian-based Unified Nearest-neighbor/Embedding Recognition is an original framework for representation matching across domains.
It explores intelligence as a search process: structured representations are decomposed, indexed, and compared through explicit correspondence discovery. Similarity is grounded in identifiable structural matches, not exclusively opaque latent embeddings.
Future research
HUNTER is the first step, not the boundary.
We are developing further frameworks, tools, and computational models for interpretable and trustworthy artificial intelligence.
Publications & open science
Research is stronger in the open.
We share work through research papers, technical reports, open-source software, demonstrations, and collaborations. Scientific rigor and a clear record of methods are central to our practice.
Research papers
Technical reports
Open-source software
Collaborative inquiry
Our vision