Semantic Memory Systems
AI systems that remember what matters.
AI-enabled systems that know what matters, what they don't know, and how to tell the difference—like you.
The Problem Nobody Wants to Admit
Your AI is lying. Not maliciously—it just doesn't know any better.
The Chatbot Problem
Your customer asks a simple question. Your chatbot answers confidently. The answer is wrong—it's from a policy you updated six months ago. The chatbot doesn't know the difference between current truth and historical artifact.
The Knowledge Base Problem
Your team searches the internal wiki. They find four documents about the same process. Each says something different. Nobody knows which one is right. The wiki doesn't know either.
The Drift Problem
You update the source document. The training materials don't update. The chatbot doesn't update. The FAQ doesn't update. Six months later, you have seven versions of "truth" and no way to know which systems are lying.
If you can't answer immediately whether your AI is lying, the answer is yes.
What is Semantic Memory?
Cognitive science distinguishes between two types of memory:
Episodic memory remembers events. When you learned something, where you were, what happened.
Semantic memory remembers meaning. You know that Paris is in France, but you don't remember the moment you learned it. The fact is just... known.
Most AI systems have only episodic memory. They store documents with timestamps. They retrieve based on similarity. They have no idea what's true—only what was stored.
Semantic Memory Systems remember what matters and forget the rest.
They don't chase perfect recall. They establish canonical truth, verify it at the source, and generate confidently from verified claims. When they don't know something, they say so.
How It Works
Three layers. One architecture.
Canonical Knowledge Base
Knowledge stored as verified claims, not documents. Each claim has an owner, evidence, review date, and derivation trail.
Governed Derivation
Documents are outputs, not sources. All content derives from canonical claims. One truth, multiple presentations for different audiences.
Discrimination Infrastructure
The system knows what it knows, what it doesn't know, and can tell the difference. AI that says "I don't know" instead of confidently lying.
Proof Points
We don't just consult—we build.
TerpTune
AI cannabis concierge that maps personal neurochemistry to terpene profiles. Karl remembers your patterns, not your timestamps. Semantic memory for personal data.
Book of Fire
Thesis on humans as semantic ordering specialists in the age of AI—built using the methodology it describes. Recursive proof that this architecture works.
Industries
Semantic Memory Systems for every vertical where truth matters.
Healthcare
Protocol says one thing. Order set says another. Patient is in the middle. We fix that.
Software
Your docs are wrong. Your users found out before you did. Documentation that stays current with code.
Retail
Everyone has the price. Nobody has the same price. Consistency across channels and locations.
Education
You know something is wrong. You just can't prove it. Curriculum alignment that's architectural, not archaeological.
About
SemanticMemorySystems is the consulting practice of Mark Ulett.
I've spent decades watching organizations build systems that don't remember correctly. Documents drift. Knowledge bases lie. AI amplifies the problem by generating confidently from wrong sources.
Semantic Memory Systems are how we fix this. Not with better search. Not with more training data. With architecture that makes truth verifiable and lies detectable.
The Book of Fire is the theory. TerpTune is the proof. The methodology is battle-tested.
If your organization stores knowledge in documents, and those documents drift, and AI is making that drift worse—we should talk.
Ready to Stop the Lying?
Your AI doesn't need more training data. It needs memory architecture—canonical truths, verified claims, and the infrastructure to say "I don't know."