The technical ground we build on.
Six areas that, together, let a system take in the world, make sense of it, and act on what it finds.
Six nodes. One living graph.
Select a domain to move through the system. Each node is a research and product surface — together they ground Karmact and everything we ship.
Connections as first-class citizens.
A knowledge graph (KG) is a programmatic model of a domain that represents real-world entities, concepts, operations, and events — and the relationships between them. Structurally, it's a graph-structured database or semantic network, organising data as nodes connected by edges, supported by a formal schema or ontology.
Unlike a traditional database that stores data in isolated tables, rows, and columns, a knowledge graph treats the connections between data points as first-class citizens — letting a system reason through context, and infer new, unstated facts.
The foundation pillars.
At its core, intelligence is the capacity to process information, learn from experience, reason through complexity, and solve novel problems. An intelligent system is built on several foundation pillars — problem solving, reasoning, adaptability, learning, and retention.
The measure of it, in practice, is a system's capacity to take in raw data from the world, extract meaning from it, and convert that meaning into deliberate action — crowd intelligence and hybrid intelligence, working together.
Finding the truth in what's already there.
Information retrieval focuses on searching, organising, and ranking unstructured or semi-structured data — text, images, web pages — from large collections, to satisfy a specific need.
Building a system this way means constantly checking where it stands, and how well it corroborates with everything else already known — which is where IR does its deepest work: finding truth at the core of a system, not just matches at the surface.
Making words point at something real.
Natural language grounding is the process of mapping linguistic symbols — words, phrases, sentences — to rich, non-linguistic representations of the physical, digital, or conceptual world. It bridges the gap between abstract syntax and semantic reality, so a system doesn't just manipulate words by statistical pattern, but actually understands what those words refer to in context.
Without grounding, a language model runs into what cognitive scientists call the symbol grounding problem: a closed loop where words are defined only by other words — like learning a language from nothing but a monolingual dictionary.
Where intelligence meets the physical world.
Networks of sensors and robotic agents extend everything above out of the graph and into physical space — gathering the raw signal a system reasons over, and carrying its decisions back out into the world.
Meaning held in layout, not just words.
Many real documents carry meaning in their structure as much as their language — tables, forms, headers, and layout. This domain focuses on reading that structure alongside the text, so nothing meaningful is lost in translation to plain text.