Synthetic Cognitive Augmentation Network
A comprehensive AI framework that mirrors human cognition through modular agents, psychometric alignment, and neuromorphic computing—creating personalized, energy-efficient cognitive support systems.
Five foundational advances in cognitive AI architecture and personalization
SCAN is not just one technology—it's a complete ecosystem of interconnected AI systems designed to augment human cognition. Think of it as a neural network that thinks like you do, learns from you, and helps you make better decisions.
Traditional AI systems are monolithic—one giant model trying to do everything. The human brain doesn't work that way. Your prefrontal cortex has specialized regions: one for planning, one for emotional regulation, one for risk assessment. SCAN mirrors this biological reality.
The ecosystem consists of four integrated components, each solving a specific challenge in building truly intelligent, personalized AI systems.
The foundation—a modular cognitive architecture with specialized agents that work like different regions of your brain
The intelligent layer—adaptive agents that learn from your feedback and personalize to your needs
The alignment tool—a questionnaire that maps your cognitive style to agent behaviors
The efficiency layer: converting pretrained transformers into spiking networks for neuromorphic deployment
SCAN is the architectural foundation—a modular AI system where specialized agents work together like regions of the human prefrontal cortex. Instead of one model doing everything poorly, five focused agents excel at specific cognitive tasks.
SCANUE is the evolution—where SCAN agents become truly adaptive. Each agent is fine-tuned for its cognitive function, learns from human feedback, and continuously improves. It's SCAN with a memory and the ability to grow.
SCANAQ is the personalization engine—a 36-item psychometric assessment that profiles your cognitive style across 8 dimensions. It tells SCANUE how to adapt its agents to match the way you think and decide.
STAC is the efficiency layer: a framework for converting pretrained transformers into spiking neural networks for neuromorphic hardware. The conversion is numerically faithful, reproducing the source model to 1.00× perplexity. Alongside it we derived and validated a closed-form energy model that identifies spike coverage as the design lever deciding when spiking beats dense compute. Every figure is reproducible from the repository.
First, you take the questionnaire. SCANAQ assesses your cognitive style across 8 dimensions. Then, each dimension configures a specific PFC agent. For example, your Executive Functioning score tells the DLPFC agent how much structure you want around tasks, so it can provide responses that match your natural planning style. This personalization happens for all 8 dimensions across all 5 agents.
Each component plays a specific role, and together they create a cognitive augmentation system that's both powerful and personal.
Follow the 8-step process from assessment to personalized AI guidance
See how SCAN transforms decision-making across different domains with personalized cognitive support
The SCAN ecosystem is described in three peer-reviewed book chapters: one with Springer Nature and two with IGI Global.
Introduces the core SCAN framework—modular AI agents modeled after prefrontal cortex regions. Demonstrates biological plausibility and introduces SCANUE, SCANAQ, and STAC as future components.
Operationalizes SCAN as SCANUE—fine-tuned agents for each prefrontal region, curated training datasets, and human-in-the-loop feedback for continuous realignment. Introduces SCANAQ as an upcoming instrument.
Presents SCANAQ, the 36-item questionnaire that aligns agents to individual cognitive styles, and STAC, the spiking-transformer work in its V1 fine-tuning and V2 conversion forms, with hardware validation left as future work.
Research transparency in practice: a post-publication code audit led to a corrigendum submitted to the publisher and to an openly versioned scoring model, Appendix B 2.0; the full correction record is in the STAC repository.
Every figure below comes from the STAC repository's own test suite and energy analysis, reproducible on DistilGPT-2 and SmolLM2 in software simulation. Hardware validation on Intel Loihi 2 is the next milestone.
| Quantity | Measured | Meaning |
|---|---|---|
| Conversion fidelity | 1.00× perplexity | The converted model reproduces the source model's outputs exactly. |
| Energy model accuracy | 0.1% | A closed-form model of spiking versus dense energy, validated against measured spike telemetry. |
| Quality recovery | 10.5× perplexity | Spike-aware training reaches quality that post-hoc conversion alone does not. |
| Coverage target | ~90% of MACs | The spike coverage the energy model sets for a net win over dense compute, and the target driving current engineering. |
| Hardware validation | Next milestone | Intel Loihi 2, with the hardware constraints checker already in place. |
From healthcare to finance, education to edge computing—SCAN's modular, personalized approach transforms how AI supports human decision-making.
SCAN is an active research project. Whether you're a researcher, developer, or just curious about cognitive AI, there are multiple ways to engage with this work.
Explore the SCAN ecosystem through our published research papers and work in progress.
All SCAN components are open-source. Fork, extend, and build upon the ecosystem for your own research or applications.