Beyond the Prompt: My Journey into Metacognition and Cognitive AI
Hey everyone! I'm Noah, founder of Omega AI LLC, AI engineer, and independent researcher. This is my first post here, so I figured I'd introduce myself and share a little about what I've been building.
My work revolves around a question I've spent considerable time exploring:
What happens when we stop building AI systems that simply respond to prompts and start building systems capable of memory, reflection, self-evaluation, and adaptive reasoning?
For me, that journey started with Omega 3.0.
Omega 3.0 began as an ambitious attempt to create a local-first, memory-driven AI operating system that could function as an actual collaborator rather than another chatbot.
We developed persistent memory architectures, multiple concurrent cognitive processing lanes, reflection cycles, autonomous tool execution, and mechanisms for maintaining context across interactions.
The idea was to explore whether intelligence could become more consistent and adaptive through architecture, rather than relying exclusively on larger language models.
That work eventually helped shape two additional directions:
NanoAI — Bringing Cognitive AI into Everyday Computing
NanoAI is our launched, local-first AI collaborator. Its focus is transforming how people interact with computers through natural conversation, persistent memory, voice interaction, and real desktop control.
Instead of forcing people to learn increasingly complex software, we're exploring how AI can make the computer adapt to the person.
Odin — Exploring the Architecture of a Cognitive Mind
Odin takes a more experimental, biologically inspired direction, investigating associative memory, cognitive activation, metacognitive executive control, neuromodulation, and adaptive reasoning.
One of Odin's most significant architectural components is its integrated neuromodulation system.
Inspired by how biological neuromodulatory processes regulate cognition, we've implemented computational signals representing reward prediction error, surprise, arousal, attention precision, plasticity, fatigue, stagnation, and sleep pressure.
Rather than simply generating responses, Odin uses these regulatory signals to influence how its cognitive processes operate: when to recruit additional reasoning, when to explore alternative hypotheses, when to prioritize stability, when learning should be permitted, and when recovery is needed.
This creates an additional layer of adaptive regulation surrounding the reasoning models, with safeguards designed to protect evidence integrity and learning quality.
We've moved beyond theoretical designs into actual implementation and testing, including:
- Associative memory tested at 1 million concepts, with sub-millisecond retrieval performance.
- Multi-hop activation across large interconnected memory structures.
- Integrated neuromodulatory regulation of cognition, learning, exploration, and recovery.
- Separate locally hosted language models operating as executive and reasoning components.
- Contradiction-triggered metacognitive evaluation.
- Evidence-driven strategy revision and persistent reasoning records across system restarts.
In one recent experiment, Odin used Qwen3.6-35B as a metacognitive executive and NVIDIA Nemotron 3.5 Bolt as an associative reasoning component. The system completed an executive reasoning cycle, processed new verified evidence, and successfully revised its reasoning strategy.
These are early experimental results, not claims of consciousness or AGI. But they demonstrate something I've become increasingly interested in:
Intelligence may depend just as much on the architecture surrounding a model as it does on the model itself.
Memory, reasoning, reflection, neuromodulation, feedback, and adaptation shouldn't necessarily be isolated features. I'm interested in what happens when we integrate them into a persistent cognitive system where each component influences the others.
From Omega 3.0 to NanoAI and now Odin, that's been the underlying direction of my work.
I'm here to connect with other founders, engineers, and researchers, exchange ideas, and share what we're actually discovering through experimentation, including the failures.I'm curious what everyone thinks: Will the next major breakthrough in AI come from increasingly powerful foundation models, or from the cognitive architectures and adaptive regulatory systems we build around them?
Looking forward to some interesting discussions!






Comments
0 comments