Rich Canopy
Threading semantics through data science,
powering true agentic workflows
Tech revolutions require whole new ways of working; too many pretend that AI data science is their old legacy product with a few tweaks bolted on.
Models need context, context needs semantics, and humans need instant feedback. We're building the modern stack.
Projects
Core workflow
Agent-friendly dataviz
Significant MS Excel extensions
Manifesto
Source code used to cost a fortune, both to write and to maintain.
The source of truth for our software systems' behaviour had to be the executable asset, which we'd paid so much to create. "Show me the code" was king.
Yet code is usually a very poor vehicle for domain modelling. Describing the semantics of our problems and systems in general purpose, executable languages brings at least an order of magnitude extra accidental complexity over what's essential.
It's clear that code will always be necessary in design; a verification test or specific (usually pure) algo you want to pin down very precisely. Natural language is never going away, too - our formalisms ultimately bottom out on intuitive concepts.
But to define what a system is and how it should behave we need a lightweight way of building new semantics, which agents can iterate on locally, and teams can interoperate with at scale. Flexibility and rapid change when the world is closed - within the project - and stability when open, with a clean flow from one to the other.
RDF, SHACL and knowledge graphs are the perfect tool, but for modelling the engineering system itself, not the data it handles. Agents have remarkable abilities to re-systematise a knowledge graph when new data arrives. When semantics are transparent and durable, code becomes a build asset, and can be regenerated at will.