What is an intelligent ontology?
An ontology describes the things in a domain and the relationships between them. In this research, an intelligent ontology is a structured memory system that connects proposed facts to their sources, relevant time periods, and supporting or opposing evidence.
The aim is to help an agent ask more than “What did I retrieve?” It should also be able to ask “What supports this claim, what conflicts with it, and does it apply to this situation?”
Why this matters for useful AI.
Business knowledge changes. Documents disagree, procedures expire, and similar names can refer to different people or assets. A useful memory system needs to preserve those distinctions instead of flattening them into a confident answer.
We are exploring a shared evidence graph for software agents, with a longer-term interest in affordable devices that cooperate on local networks. Local operation and constrained hardware are design goals; their practical value still needs to be measured.
What the current prototype does.
The delivered fact-review tool checks proposed facts against a synthetic evidence graph. It retrieves supporting and opposing evidence with citations, handles time and identity distinctions, and records human review separately from its automated evidence result.
- Evidence stays attached. A retrieved conclusion can be inspected alongside the sources and relationships used to reach it.
- Disagreement stays visible. Supporting and opposing evidence are considered explicitly rather than silently merged.
- Context matters. Time, identity, and relationship paths help distinguish a relevant fact from a misleading match.
- Review is a separate action. A human review record does not silently insert a fact into the graph or authorize a downstream tool.
The current demonstration uses synthetic evidence and deterministic checks. A proposed-fact comparison pilot has been prepared, but no model comparison has run. This work has not yet demonstrated reliable commercial or field-swarm value.
It is a research prototype, not a generally available product. Production performance, customer benefit, and advantages over existing memory approaches remain questions to evaluate.
Where the research is headed.
The immediate direction is graph memory for software agents: a local document-and-task workflow that makes retrieval, evidence, and review visible. A later evaluation will test whether a model adds useful value over simpler methods.
A longer-term mission is affordable technology for soil health and food production. That work requires appropriate field and laboratory evidence, relevant context, and measured outcomes. The current software prototype does not establish agricultural performance.
Research and delivery, with clear boundaries.
Kelly Green Holdings develops this research alongside its commercial portfolio. Starbloom Consulting delivers private AI and workflow automation projects today; the ontology research is a separate effort.
For research partnerships, contact hello@kellygreenholdings.com. For a business automation project, start a conversation with Starbloom.
Research led by Darren Kelly, founder of Kelly Green Holdings.