PRD-LLM v2.1 : A Complete Unified Framework for Truth-First Causal Artificial General Intelligence

This repository contains the complete source code and implementation of PRD-LLM (Pattana-Relational Dynamics Large Language Model), a novel architecture for truth-first causal reasoning. The system is based on the 2-2-1-2-1 layered design and Relational Tensor representation [C, W, L, T, U, D], integrating Myo Min Aung Unified Theory (MUT) equations with recursive feedback loops to enable explainable AI outputs and robust self-correction. The included code provides:- Core Relational Tensor implementation with MUT equation integration- 2-2-1-2-1 layered architecture (Perception, Relational Core, Verification, Synthesis)- Recursive feedback mechanism for self-correction (37% uncertainty reduction)- Dataset generation pipeline using free LLMs (DeepSeek-R1 via Ollama)- Complete PRD-LLM engine with conflict resolution and temporal decay- Training scripts with hyperparameter optimization- Batch testing and visualization tools- CPU deployment with quantization and FastAPI server- Docker configuration for containerized deployment All components are implemented in Python and designed to run locally using open-source tools (Ollama, MetaGPT). No API keys or internet connection required after initial setup. Key features:- Truth-first causal reasoning with explicit uncertainty handling- Self-correcting via recursive feedback- Explainable outputs with full traceability- Conflict resolution using scoring function S(R) = αC + βW + γ(1-U) + δT- Temporal dynamics with exponential decay T(t) = T0·e^(-λΔt)- Compatibility with SU(5) relational algebra This repository accompanies the paper: "PRD-LLM: A Complete Unified Framework for Truth-First Causal Artificial General Intelligence" (available in the main PDF).