ChineseHarm-Bench: A Chinese Harmful Content Detection Benchmark

[ACL 2026] KnowRL: Exploring Knowledgeable Reinforcement Learning for Factuality

[ACL 2026] Executable Knowledge Graphs for Replicating AI Research

[EMNLP 2025 Findings] ClueAnchor: Clue-Anchored Knowledge Reasoning Exploration and Optimization for Retrieval-Augmented Generation

[ICML 2026 Spotlight] Know More, Know Clearer: A Meta-Cognitive Framework for Knowledge Augmentation in Large Language Models

Executable, API-key-free lab for RAG, GraphRAG, KAG, CAG, memory, table, and tool augmentation

HcTAG (human-centric Thinking-Augmented Generation) — A prompt augmentation framework for guiding LLMs with layered, human-driven thinking before generation.

Self-audited knowledge augmentation pipeline for DeepSeek V4 Pro. The model maps its own training gaps, researches them from official sources, and consumes the result as on-demand skill documents — closing the cutoff gap on fast-moving TypeScript stacks.

Prototype for retrieval-augmented generation (RAG) with vector DBs and open-weight LLMs. Documents early experiments, design ideas, and multimodal model directions.