Query-adaptive WRRF methods for production search retrieval, showing optimal fusion weights form intervals rather than points and explaining RRF's near-optimality for a large share of queries.

An agentic system that iteratively synthesizes and refines formal specifications, going beyond verifier acceptance to ensure genuine correctness and completeness.

Extracting full books from Llama models via prefix prompting, and showing that instruction-tuning can undo alignment's memorization mitigations.

A RAG-based conversational assistant for Congenital Diaphragmatic Hernia questions, with accuracy improving substantially after knowledge-base expansion.

RAGMan is an LLM-powered tutoring system that can support a variety of course-specific and homework-specific AI tutors.

LLM shows “great promise” in code synthesis. Can LLM keep "the promise" to ensure that its synthesis code is formally correct?

Evaluating ChatGPT against prior automated approaches for commit message generation, finding it substantially outperforms methods trained specifically on commit data.

The research focuses on enhancing the quality of commit messages in collaborative software development by introducing specific requirements in the "what" and "why" parts through modifications to the GitHub Desktop application, demonstrating improved informativeness, clearness, and length in commit messages among students.