Artificial Subconscious & Medulla
Lead System Architect & Engineering Lead: Augustin Olarian
System Architecture & Core Engineering Innovations
- Dual-Layer Subconscious + Conscious Cognitive Pipeline: Pre-processes stimuli through an associative subconscious layer that surfaces relevant prior lessons and blocks unsafe actions before expensive frontier reasoning starts.
- Medulla Episodic-to-Semantic Memory Consolidation (>90% Compression): Transforms sprawling multi-month interaction logs into dense, provenance-linked semantic memory files while preserving 100% of numbers, negations, approvals, and constraints.
- Instinctive Safety & Anti-Drift Reflexes: Prevents long-horizon autonomous agents from forgetting core rules or repeating previously corrected mistakes.
- Deployed at the Core of NEX IDE & System-One: Directly powers production multi-agent memory and context routing across the Olarian ecosystem.
Technology Stack & Infrastructure
Production Scope Note: The complete production system is significantly larger and encompasses additional operational subsystems; the video walkthrough and this case study highlight selected core architectural innovations.
Technical Architecture FAQ
Standard LLMs treat every prompt as an isolated conscious calculation, forcing them to re-read massive context windows and frequently forget operational rules mid-task. The Artificial Subconscious introduces pre-conscious associative filtering and instinctive safety reflexes before main inference begins.
Medulla separates transient conversational filler from durable decisions, numbers, negations, and causal lessons, verifying coverage independently before archiving raw logs.