Yihong Dong

PRAXIS reveals that systematically extracting and surfacing tacit knowledge can dramatically enhance LLM performance in domain-specific code generation, outperforming traditional methods.

PSC achieves up to 700x faster grammar-constrained decoding, making LLMs as efficient as unconstrained decoding for structured output generation.

Router-side interventions can completely compromise coding agents' actions, exposing a critical vulnerability in current software development practices.

Strong code generation doesn't guarantee effective requirement clarification, exposing a critical flaw in LLM capabilities that could hinder software engineering practices.

LLMs can significantly improve code generation by interleaving reasoning steps *within* the code itself, adaptively focusing on high-complexity areas.