Engram - Cognitive Mesh Architecture

Author: Vincent Kaufmann Affiliation: Engram Research, Dubai, UAE Date: February 2026

Overview

Engram is a cognitive architecture that replaces autoregressive token prediction with iterative diffusion-based reasoning, biological state modulation, and associative memory.

Core components:

• Cognitive Mesh - Modular neural interconnect with specialized nuclei
• Biological Architecture - 72-dim internal state engine with FiLM conditioning
• Associative Memory - Persistent episodic recall with reflexive retrieval
• Grid Diffusion Transformer (GDiT) - Discrete diffusion for spatial reasoning
• World Model - Inner diffusion for mental simulation and hypothesis testing

Files

• pitch-deck.html - Interactive pitch deck (open in browser, arrow keys to navigate)
• Engram-Pitch-Deck-HiRes.pdf - PDF version of pitch deck

Citation

Kaufmann, V. (2026). Engram V3: A Cognitive Mesh Architecture for Abstract Reasoning. Engram Research.

License

This work is licensed under CC BY-NC 4.0. See LICENSE.