Publications

• The Genealogy of Ideology: Predicting Agreement and Persuasive Memes in the U.S. Courts of Appeals

  Opinion length, citation counts, and judges’ voting histories offer clues to when a panel will disagree.

• Counterexamples as Feedback for Agent Self-Correction

  On regex tasks, tests whether showing a failing input helps more than simply asking an agent to try again.

• DocSync: Agentic Documentation Maintenance via Critic-Guided Reflexion

  Gives a small language model code structure and retrieved context to check proposed edits against the source.

• Transformer-Based Predictive Maintenance for Risk-Aware Instrument Calibration

  Compares fixed schedules with interventions guided by sensor histories, forecast uncertainty, and the cost of missed violations.

• Scalable Sequential Recommendation under Latency and Memory Constraints

  Packs item attributes into fixed-size vectors and processes user histories in linear time as sequences grow.

Google Scholar