graph-reasoning

[KDD 2025] Rewarding Graph Reasoning Process makes LLMs more Generalized Reasoners

Publishable notebook/reference pipeline for evidence-grounded conflict analysis over knowledge graphs.

Agent Skill for Claude: Build, query, validate, and reason over knowledge graphs. Adversarial fact validation, deterministic reasoning, JSON-LD schema.

Persistent semantic state engine for AI agents — knowledge graphs, dynamic RAG, semantic zoom, bidirectional natural language translation, provenance tracking, self-extending graph.

Graph-based RAG autonomous agent with dynamic task decomposition, multi-LLM support, and Ragas evaluation

Deterministic temporal and relational reasoning engine with historical state, contradictions, provenance, exact multi-hop reasoning, and reproducible benchmarks.

ARIA - Adaptive Revenue Intelligence & Action. An ATLAS-class payment revenue-recovery system: maps payment dependencies as a graph, observes degradation, traces failures to root cause with evidence, selects bounded recovery actions, and measures recovered revenue against a graph-blind baseline. Deterministic, explainable, honestly evaluated.

Typed, inspectable reasoning framework for connecting candidate ideas, evidence, peer review, and evaluation.

Verifier-backed abstraction invention for small formal protocol and concurrency systems with contradiction-driven ontology revision

PyTorch implementation of SPIN Road Mapper: road segmentation from aerial images using spatial and interaction space graph reasoning on stacked hourglass networks.

Исследовательский код для сегментации и структурного продолжения тонких сейсмических разломов с использованием анизотропной геометрии, топологических ограничений и графового анализа LIRA.

Knowledge Representation & Reasoning via AIML, Pytholog, and Neo4j Graph DB

Evidence-grounded recall tracing with LLM-assisted extraction, deterministic graph analysis, and auditable human review.

Self-reflective, hallucination-aware multimodal RAG for reliable vision-language reasoning.

A Proposer-Critic debate framework testing whether active verification reduces the encoding-fragility of LLM graph reasoning, benchmarked against a matched-compute majority vote on GraphQA. TAU coursework.