I’m Hercules , a builder at heart — part systems architect, part engineer, part entrepreneur.
My work lives at the intersection of AI architecture, decision systems, and real-world execution. I’m less interested in demos and more interested in systems that survive contact with reality: messy data, imperfect signals, cost constraints, and human behavior.
Over the past years, I’ve been designing and implementing hybrid AI systems that combine:
-
Large Language Models (LLMs)
-
Small / specialized models (SLMs)
-
Knowledge Graphs
-
Deterministic rules and outcome-based reasoning
The goal is simple but hard:
build AI that knows when to act, when to stay silent, and when not to pretend it knows.
The Cybertron Architecture
Cybertron is the internal name for the architecture I’m building.
It is not a single model, and it is not “LLM-first”.
Cybertron is a layered cognitive system, inspired by how humans actually make decisions:
-
LLM for language, abstraction, and hypothesis generation
-
SLM / rule engines for precision, stability, and cost control
-
Knowledge Graph (KG) as long-term structured memory
-
Outcome / consequence models to evaluate actions before execution
-
Agent orchestration to let components collaborate, disagree, and self-correct
In Cybertron, an LLM is never trusted alone.
Every answer is shaped, constrained, or vetoed by structure, memory, and consequences.