About CogVexa
Our mission is to build practical AI systems that improve human decisions.
CogVexa is an applied AI company focused on reliability, transparency, and real-world product outcomes.
We publish and open-source research where it helps the field, and we ship products—Simfolk for behavioral twin cohorts, DuckAI for an OCR sidebar copilot—so ideas turn into systems teams can run. Milestones and dates live on News.
Our vision
Make every human digital persona. We are building toward a world where each person can have a trustworthy, evolving digital representation that supports better reasoning, clearer context, and higher quality decisions across product, research, and operations.
Our long-term direction is to make digital personas practical and reliable at scale: privacy-aware, auditable, and useful in production. We focus on making intelligence legible so teams can challenge outputs, improve them continuously, and confidently ship real outcomes.
That vision shows up in Simfolk as twins tied to real humans—inspectable memory, cohort debate, and auditable simulation. DuckAI stays a distinct surface: screen context via OCR and tool integrations, without mixing twin cohorts into a generic chat box.
Our journey
We published NCG (Novelty-triggered Capacity Growth): novelty-triggered capacity expansion with meta-parameters, strong Split-MNIST and Split-CIFAR-10 forgetting reductions, and honest reporting where meta-parameter recovery is not always clean. ARIA (Adaptive Reasoning with Integrated Architecture)—including the ARIA-Ω variant—is more experimental and currently deprioritized relative to NCG.
Alongside research, we ship Simfolk for cohorts of behavioral human twins and DuckAI for OCR-based sidebar assistance—different problems, different interfaces.
ARIA explores sparse content routing, cortical-column-style modularity, and Hebbian synaptic memory; aria-torch exists but is less mature and less rigorously validated than ncg-torch. Those research threads inform how we think about evolving models and modular reasoning—not always as literal product features.
Research does not sit in a silo: it informs what we dare promise in production, what we measure, and how we prioritize Simfolk versus DuckAI. For paper details and PDFs, see Research.