A Hybrid Cognitive Architecture for Symbolic Reasoning in Large Language Models
Abstract This research note presents The Ghost Conductor Framework, a hybrid cognitive architecture designed to separate language generation from symbolic reasoning in large language models.The system routes natural language through a narrative module while delegating formal logic and mathematical inference to a dedicated reasoning engine. A structured Memory Vault provides long-term continuity, integrating DeepSeek-10× text compression for efficient archival.Initial experiments show significant reduction of hallucinations and improved logical accuracy in multi-step tasks, demonstrating that consistent reasoning can be achieved without retrai
Resource information
- Authors
- Vinte Marinescu, Gabriela Diana
- Research area
- Artificial intelligence
- Research topics
- large language models, LLM reasoning, artificial intelligence, hybrid reasoning, symbolic logic, cognitive architecture, neural-symbolic integration, autonomous agents
- Publication date
- 2025
- DOI
- 10.5281/zenodo.17423216
- Upstream last modified
- Oct 23, 2025
- Last record update
- Nov 5, 2026
About this resource
Abstract This research note presents The Ghost Conductor Framework, a hybrid cognitive architecture designed to separate language generation from symbolic reasoning in large language models.The system routes natural language through a narrative module while delegating formal logic and mathematical inference to a dedicated reasoning engine. A structured Memory Vault provides long-term continuity, integrating DeepSeek-10× text compression for efficient archival.Initial experiments show significant reduction of hallucinations and improved logical accuracy in multi-step tasks, demonstrating that consistent reasoning can be achieved without retraining the base model.Developed independently by Ardent Robotics (Diana Gabriela Vinte Marinescu), this work proposes a practical pathway toward embodied, introspective AI systems that think before they speak.