PatchWorld: Learning Executable World Models Without Gradients

We introduce PatchWorld, a gradient-free framework that induces inspectable, replayable, and locally repairable Python world models from offline agent trajectories for prediction and planning.

Automating the Grounding Layer: A Survey of Operational Knowledge for Enterprise Agents

Enterprise agents need more than RAG. Our survey maps how LLMs can build the structured grounding layer—vocabulary, constraints, actions, governance, and maintenance—that makes agents actually deployable.

The Agent Harness: Infrastructure That Makes LLM Agents Actually Work

Led by Zhongwei Xie, our survey maps the four-layer Agent Harness stack—execution, context, interaction, and guardrails—and why it may matter as much as the model itself.

Invited talk at Home Depot on knowledge graph construction

A short overview of my Home Depot talk on building knowledge graphs, from rich semantics to autonomous construction and task-aware optimization.

Three papers accepted to ACL 2026

AutoSchemaKG, AutoGraph R1, and multimodal abductive reasoning—three angles on building and using knowledge for language models.

From Knowledge Graphs to Neural Graph Databases: Data, Algorithms, and Applications

Talk on how we move from symbolic knowledge graphs to neural graph databases, and what this means for data, algorithms, and real-world applications.

[ICLR'26] Taming the Wild Hypothesis: Introducing Controllable Abductive Reasoning in Knowledge Graphs