Latest Advances on System-2 Reasoning

Implementation of Discrete Key / Value Bottleneck, in Pytorch

A comprehensive implementation of a Neurosymbolic framework for Visual Question Answering (VQA) on the CLEVR dataset. This project translates natural language questions into symbolic programs using three different learning strategies: Supervised (LSTM & Transformer), Reinforcement Learning (REINFORCE), and In-Context Learning (LLM).

An exploration of AI-driven symbolic regression, implementing and analyzing two key methods: Equation Learner (EQL) neural networks and a Seq2Seq Transformer model.

System 2 Reasoning for LLMs. An inference-time Supervisor Kernel that forces verification before generation to mitigate hallucinations. "Latency is the price of Integrity."

An implementation of the GraphRAG pipeline (based on the 2024 paper "From Local to Global" by Edge et al.) for query-focused summarization of large text corpora.

A project exploring agentic AI for Visual Question Answering. Compares zero-shot VLM performance (Transformers) against advanced pipelines using OpenCV (tool-use) and Chain-of-Thought (reasoning).

CRSM (Continuous Reasoning State Model): An asynchronous "System 2" architecture that implements Hierarchical State Sovereignty within a Mamba backbone. Unlike traditional search wrappers, CRSM uses Forward-Projected Planning and Sparse-Gated Injection to steer latent manifolds in real-time, decoupling strategic reasoning from token generation.