Research
We publish what we build, open-source what we can, and treat peer review as a quality requirement — not a marketing asset.
Why we publish
The architecture is not a secret.
The implementation work is
Most enterprise AI vendors treat their architecture as proprietary differentiation. We take the opposite view: the architectural ideas behind a trustworthy AI system should be inspectable, citable, and challengeable by other researchers. The defensible part of an enterprise offering is not the idea — it is the engineering, the integration, and the consulting work that turns the idea into a deployable system.
That orientation has three practical consequences:
The contract layer, the knowledge graph construction architecture, and the symbolic engine that ties them together are described in peer-reviewed and pre-print literature. Anyone evaluating our claims can read the source material directly.
Our framework is open source
The neurosymbolic engine — SymbolicAI — is released under BSD-3 and developed in the open. The code that implements the contracts described in our papers is the same code that runs in production deployments.
Our research is informed by — and where appropriate, conducted with — the broader academic community in neurosymbolic AI, knowledge representation, and formal methods.
Research focus areas
What our team works on
Knowledge graph construction and reasoning
Hybrid approaches in which language models are used to extract entities and relations under symbolic constraints, with the symbolic layer enforcing schema, type, and referential conditions during construction. The aim is knowledge graphs that carry their own provenance and can be queried as constraint satisfaction problems, not only as pattern-matching targets.
Verifiable retrieval and synthesis
Retrieval architectures that produce outputs whose claims can be mechanically traced back to source evidence, with formal guarantees about the relationship between retrieved passages, generated content, and cited support. This work sits at the intersection of information retrieval, formal methods, and language-model-based generation.
Open-source infrastructure for neurosymbolic AI
The engineering work required to make neurosymbolic primitives — contracts, validators, structured prompts, symbolic operators — usable as a framework rather than as research artifacts. This is the work behind SymbolicAI, and it continues as the framework matures.
Selected publications
Read what we publish. Cite what's useful.
The technology described on this page is grounded in published research. The relevant work:
Trustworthy Agent Design
A practical whitepaper on designing trustworthy LLM agents with contract-based controls that validate inputs, outputs, and semantic requirements before agents act.
HyDRA - Knowledge Graph Construction
A whitepaper on HyDRA, a hybrid-driven reasoning architecture that uses collaborative agents, competency questions, and verifiable contracts to automate reliable knowledge graph construction.
SymbolicAI Framework (Open Source)
A developer-focused whitepaper on SymbolicAI, the open-source neurosymbolic framework for composing LLMs with Python-native symbolic abstractions, semantic primitives, and contract validation.
A complete publication list is available on request. We update this page when new work is released.
Open source
The engine, in public
The neurosymbolic engine that our research describes is released as the open-source SymbolicAI framework. It is the same code that powers the deployments we ship — there is no separate "production fork" with proprietary patches.
The framework is released under BSD-3, which permits commercial use, modification, and redistribution with minimal obligations. The repository includes the core runtime, contract primitives, symbolic operators, integration adapters for common LLM backends, and the test suite.
For organisations evaluating our work: the most efficient way to verify what is described on our other pages is to read the relevant module in the repository. The architectural claims and the implementation are intentionally close.
SymbolicAI framework
Repository:
Licence:
BSD-3-Clause
Status:
Actively maintained
Research collaboration
How we work with the wider community
We treat the research community as a peer audience, not a marketing channel. In practice, that means:
1.We respond to citations and follow-up questions
From researchers working in adjacent areas — neurosymbolic AI, formal verification of generative systems, knowledge graph construction, contract-based programming. If you are working on related problems and want to compare notes, the contact path is the same as for commercial inquiries.
2.We collaborate selectively
With academic groups whose work intersects with ours. These collaborations have produced joint publications and open-source contributions; we do not list ongoing collaborations on this page, both for confidentiality reasons and because the list changes.
3.We hire from the research community
Our team is composed of researchers and engineers with backgrounds in machine learning, knowledge representation, and formal methods. Vacancies, when they are open, are listed through standard channels — not on this page.
