Reasoning

Semantica is a Python framework for building AI systems with full explainability, auditability, and provenance tracking through knowledge graphs,...

This repository is the official code companion to a Manning book that teaches how to build a reasoning LLM from scratch using PyTorch, starting with...

SDPO (Self-Distilled Policy Optimization) is a reinforcement learning framework for post-training large language models on verifiable tasks like code...

This is a curated research paper collection focused on AI agents, organized by capabilities (reasoning, planning, tool use, memory) and applications...

PageIndex is a reasoning-based RAG system that eliminates vector databases and chunking by building hierarchical tree indexes over long documents and...

Haft is a governance substrate for AI-assisted software delivery that layers formal specification, decision tracking, and evidence management on top...

WFGY is an open-source ecosystem for AI reasoning, retrieval-augmented generation (RAG), and agent workflows, currently advancing toward version 5.0...

TypeDB is a next-generation database that combines relational, document, and graph database capabilities through a modern type system and TypeQL...

PySpur is a visual development platform for building and iterating on AI agent workflows, targeting AI engineers who need to debug, test, and deploy...

Protege Desktop is a free, open-source ontology editor implementing the OWL 2.0 standard for knowledge graph and semantic web development. It serves...

TTRL is a research framework for applying reinforcement learning at inference time on language models without ground-truth labels, targeting...

This is a curated survey and collection of papers on reinforcement learning techniques applied to large reasoning models (LRMs), organized as an...

Skywork-R1V is a series of advanced multimodal reasoning models that combine vision and language understanding with chain-of-thought capabilities,...

HRM is a specialized research model for complex reasoning tasks, inspired by hierarchical brain processing. It combines two recurrent modules for...

This is a research project analyzing the training methodology behind R1-Zero-like language models, with a critical examination of base models and...