Firms that grow beyond their founding team face an irreducible problem: the judgement that made them exceptional lives in a small number of heads. We build the infrastructure that makes it transferable, scalable, and resilient to the inevitable mobility of people.
Every firm that grows beyond its founding team confronts the same structural problem. The Rolodex stays in the CRM. The analyses stay in the archive. What leaves is the mental model — the taxonomy of situations, the pattern library, the sense of what matters and why, the way of reading a client or a market before the numbers confirm it.
Standard RAG systems make the archive more accessible. They do not make the firm's accumulated interpretive framework transmissible. An AI that has read your documents reasons from the market average. An AI that has internalised your framework reasons like you.
Merlin Intelligence builds the knowledge infrastructure that transforms a firm's accumulated expertise into a governed, living system that AI can reason over — not just retrieve from.
Documents, analyses, client materials, voice transcripts. The raw material of everything else. Without it, there is nothing to reason over. With it alone, there is only retrieval.
The causal structure of the domain — what leads to what, under which conditions, in which regimes. The layer that transforms a generic LLM into a firm-specific reasoning system.
Experiential knowledge captured continuously, feeding the ontology from the inside. The layer that ensures the system never stops growing — and never stops being yours.
Ontologies are the threads that stitch together the different fabrics of an organisation — contexts, teams, generations, geographies — and allow a wave of strategic coherence to propagate across the whole firm without passing through the hierarchy. A distributed C2 system where command is shared meaning and control is encoded in the structure of knowledge itself.
Our open-source semantic intelligence engine. Built for organisations that need to move beyond standard retrieval and into structured reasoning — without dependency on closed AI infrastructure.
eigenmind detects the concepts a corpus keeps returning to from different directions — the structurally unusual, informationally dense nodes that standard search misses. It extracts ontologies from text, weights edges by semantic proximity, and surfaces the knowledge singularities that precede change: in markets, in organisations, in domains. Delivered as transferable source code under MIT licence.
github.com/merlin-intelligence/eigenmind →Most of what determines whether an organisation thrives or is blindsided does not live in its dominant narrative. It lives at the margins — in the small number of concepts that keep reappearing from unrelated directions, that connect domains nobody thought were connected, that carry disproportionate structural weight relative to how often they are mentioned. We call these singularities: the relevant-but-not-obvious nodes of a knowledge structure.
A singularity is not a prediction. It is a structural property — computable, auditable, and falsifiable — of the graph that represents what an organisation knows.
Documents, transcripts and notes embedded and connected by semantic proximity. Singularities here are nodes that bridge otherwise disconnected clusters of meaning — ideas that keep surfacing across unrelated conversations without ever being named as a single concept.
Typed entities and causal relations — what triggers what, under which regime. Singularities here are causal pivots: nodes whose position governs how a shock propagates through the rest of the structure, often invisible until the regime that activates them occurs.
The union graph carries both semantic proximity and causal typing on its edges. Singularities computed here are the strongest signal class: structurally marginal in the corpus, yet causally central in the domain model — exactly the combination that precedes regime shifts.
Computing singularities is a graph-theoretic problem. We use spectral methods — the eigenvectors of the graph Laplacian decompose a corpus into clusters at multiple scales — combined with information-centrality measures such as the Lovász number and ℓ∞ bottleneck geometry, which identify nodes that are simultaneously marginal in degree and critical in connectivity. In financial terms: the events that connect markets which are not normally correlated, before the correlation becomes obvious to everyone.
This is what makes the approach relevant to two concepts that have become central to risk thinking over the last two decades: the Black Swan — a high-impact event so far outside the model that it was, by construction, unforeseeable — and the Grey Rhino — a high-probability, high-impact threat that is visible, well-documented, and persistently ignored until it is unavoidable.
Eigenmind does not claim to predict Black Swans — by definition, a genuine Black Swan sits outside any model, including this one. What it does is shrink the space of events that feel like Black Swans but are, on inspection, Grey Rhinos that the organisation's own knowledge already contained the means to see. A singularity that has been visible in the graph for months, connecting domains an organisation already has expertise in, is a Grey Rhino made computable — and therefore actionable, while there is still time to act.
MI works alongside organisations as an embedded AI and knowledge R&D partner. We transfer capability iteratively — code, know-how, and methodology — rather than producing reports. At the end of an engagement, the client owns the ontology, the source code, and the intellectual architecture. Not a dependency.