Juan Zhai

Juan Zhai is an Assistant Professor in the Manning College of Information and Computer Sciences (CICS) at the University of Massachusetts Amherst. She co-directs the Laboratory for Advanced Software Engineering Research (LASER) and is a member of the UMass NLP group. Her research advances software engineering through automated techniques for building high-quality systems with emphasis on behavioral specifications, AI safety, and trustworthy AI.

Her work addresses the fundamental challenge of aligning software behavior with intended specifications through two main directions: automated specification synthesis (translating natural language comments to formal specifications via tools like C2S and LLMCup) and defect detection/repair (developing frameworks for AI system testing, bias mitigation, and training diagnostics). Her vision integrates these into end-to-end assurance systems that continuously validate, repair, and audit evolving software in dynamic environments.

Recent publications (2024-2025) reveal dominant trends at the software engineering/AI intersection: formal specification synthesis for IoT and code generation, comment maintenance using LLMs, deep learning framework testing (DevMuT, Citadel), bias detection in LLMs, and automated training repair (AutoTrainer, DREAM). These contributions appear in top venues including ICSE, FSE, ASE, ISSTA, and ACL.

Professor Zhai currently advises PhD student Gehao Zhang (focusing on Software Engineering and AI Safety) and actively recruits new PhD/Master's students. Her LASER lab develops practical tools for specification inference, LLM-driven synthesis, and trustworthy AI, while collaborating with the UMass NLP group on language-centric software analysis.

The LASER lab, co-directed by Zhai, pioneers techniques for behavioral specification enforcement across traditional and AI-powered systems. Key projects include CPC for bidirectional code-comment analysis, ModelMeta for deep learning framework testing, and frameworks for bias mitigation across the ML lifecycle. The lab emphasizes practical, scalable tools that enhance correctness, robustness, and fairness in critical AI applications.